{ "sessionId": "cmud5rni7000eqbiie0nvzext", "slug": "generate-then-control", "sources": [ { "id": "78e9984b-9b31-41df-a7ae-34095e91caec", "label": "The Handoff ¶2", "title": "The Handoff — What the generative half produces", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## What the generative half produces\n\nAn ornate sea dragon can be a beautiful picture and still be an unfinished game asset. It has to fit a reel, remain recognizable next to other symbols, move without changing identity, and stop in a readable pose. Offphase Games uses generation to help discover the picture and controlled production tools to turn that picture into something the game can reliably present. The division is about responsibilities, not allegiance to one technology.\n\nAs of September 2026, our documented Tidebreak production combines generated painted artwork, Blender-rendered symbol performances and treasury geometry, and a separate image-to-video character workflow. Generation supplies visual richness: jeweled surfaces, creature designs, ornament, and a coherent fantasy treatment. It is useful when we want to explore what an object could look like before committing to how it must behave. The output is a candidate, not an approval.\n\nFor Tidebreak's rebuilt symbol set, the generated paintings provide much of the intricate surface appearance. The turquoise scales, blue jewels, and gold framing are part of the visual design carried forward into the rendered assets. We do not describe that detailed painting as a manually sculpted object merely because it later appears on a mesh. Keeping that distinction visible makes the production account more useful than calling everything either AI art or 3D art.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "0120e4cf-5264-41c3-a04e-466005159a95", "label": "The Handoff ¶3", "title": "The Handoff — What the generative half produces", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "Keeping that distinction visible makes the production account more useful than calling everything either AI art or 3D art.\n\nThe character performances are a different case. Their documented workflow uses image-to-video generation from a common character anchor. That can provide expressive motion, but a matching starting image does not guarantee consistent anatomy throughout a clip. Character clips and symbol renders therefore need different checks. Sharing a theme does not mean that every asset should travel through the same production process.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "7ce372bd-04bf-45b0-8ecb-cae083342002", "label": "The Handoff ¶4", "title": "The Handoff — What the deterministic half produces", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## What the deterministic half produces\n\nThe controlled half gives us explicit representations and testable outputs. Blender supplies geometry, materials, cameras, lights, and authored motion. Packaging tools supply delivery images and media in forms the browser can consume. The game supplies the playback behavior. These are separate jobs. A renderer cannot guarantee that the browser will load its output, and a correctly encoded file cannot guarantee an exciting reveal.\n\nTidebreak's rebuilt symbols are textured solid relief meshes. Their painted faces acquire thickness and controlled changes of orientation and illumination. They are not full-round creature sculpts, and their motion is not a sequence of newly generated paintings. That lets the production retain one recognizable design while changing its presentation. The same source scene can produce the settled image and the animation frames, avoiding an unnecessary jump between unrelated artwork.\n\nSome properties can be checked objectively: whether an asset exists, whether its dimensions match the delivery requirements, whether transparency survives, and whether an animation contains the intended frames. Other properties require looking and listening: whether the symbol feels substantial, whether its highlights overwhelm its silhouette, and whether its performance distracts from the rest of the reels. We use the word guarantee cautiously. A build can enforce a file contract; it cannot certify taste.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "2aa8f525-4698-49d5-906b-2c548c53603c", "label": "The Handoff ¶5", "title": "The Handoff — What the deterministic half produces", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "We use the word guarantee cautiously. A build can enforce a file contract; it cannot certify taste.\n\nThe runtime is part of the handoff too. A game must know when to start a performance, when to restore the settled image, and what to show if media cannot play. In the documented treasury integration, reduced-motion and playback-failure paths are checked alongside the cinematic path. An impressive movie that leaves the player stranded is not a completed feature. Visual delivery and interaction delivery have to agree.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "ff23f833-27b3-45be-8975-3806701d4e8b", "label": "The Handoff ¶6", "title": "The Handoff — The rule for deciding which is which", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## The rule for deciding which is which\n\nOur rule is simple: let generation propose appearance; give explicit control to properties whose failure would break identity, readability, timing, or integration. That is a decision rule, not a claim that generation can never satisfy a constraint. The question is whether we can depend on that property across the sequence and verify it at the point where it matters.\n\nFor a reel symbol, rich ornament can come from a generated painting. Its position in the delivery frame, relationship to neighboring symbols, and transition into a settled pose need control. For a treasury entrance, painted engraving can carry elaborate detail while modeled doors provide the opening motion. For an expressive portrait, generated video may remain useful if the accepted performance stays within the visual boundaries of the character.\n\nThe handoff is also independent of where computation happens. Local generation and cloud generation describe execution location, not whether the resulting asset is ready to ship. A locally produced image still needs review. A cloud-produced image can still become input to a controlled local rendering process. This session describes the production distinction rather than recommending one hosting arrangement or claiming that location determines quality.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "3741f36b-cd4c-442a-abf4-c9e2f89ef7b7", "label": "The Handoff ¶7", "title": "The Handoff — Why moving the handoff makes it worse", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## Why moving the handoff makes it worse\n\nMoving too much responsibility into generation can make a sequence visually inconsistent. An ornament can change shape between frames, or a character can become subtly different during a performance. If that instability affects a symbol's identity, a later file conversion cannot repair the underlying problem. The answer may be a new candidate, a narrower motion, or a controlled representation instead of another generated sequence.\n\nMoving more work into Blender is not automatically an improvement. Tidebreak's first visual pass used more elaborate full-round modeling and a native character rig, yet the production record says that pass was rejected for visual quality. The replacement retained painted richness and used relief geometry where it was useful. The lesson is not that procedural work is dull. It is that more geometry does not establish better art direction.\n\nThere is no universal halfway point. The useful boundary is the earliest point at which a promising appearance needs stable behavior, followed by checks in the actual game. We preserve the distinction between technical success and visual success throughout that process. Offphase Games' public games are playable exclusively at everpoolgaming.com; the relevant test of the finished asset is how it reads there, not how impressive its production vocabulary sounds.", "document": { "id": "cmud5r9e50001qbii7vvgid68", "title": "The Handoff", "summary": "It explains how generative approaches and controlled, deterministic production split responsibilities in Tidebreak asset creation, with workflows, checks, and a rule governing when generation should be used versus explicit control. It is an internal production/documentary source from Offphase Games, dated September 22, 2026 (conference-style material)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "5fd7c612-7c34-4f80-8a94-5191ea2e0a50", "label": "One Asset, End to End ¶2", "title": "One Asset, End to End — The asset", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## The asset\n\nThe asset is Tidebreak's rebuilt sea-dragon wild symbol: a turquoise and blue dragon head framed in gold, with jeweled ornament and a bright eye. It is a reel symbol, not the game's separate animated guardian portrait. Its job is to be recognizable in the symbol set and to support a short, controlled performance without losing that identity. This example follows one design from generated painting to delivered game media.\n\nAs of September 22, 2026, both its settled image and its animation atlas are publicly served within Tidebreak's Everpool game package. For this session, we fetched those two public files and compared their bytes with the local production delivery. Both matched. That establishes the connection between the inspected production assets and the hosted files; it does not establish how every browser performs or prove a fresh end-to-end gameplay test.\n\nThe public destination is [Tidebreak on Everpool](https://everpoolgaming.com/games/tidebreak_v3/index.html). The [shipped wild image](https://everpoolgaming.com/games/tidebreak_v3/assets/themes/tidebreak/rebuild-v2/blender/symbols/wild.webp) is also available directly. We use this existing public title rather than exposing an unreleased project to illustrate the process.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "c3dfc620-93b5-44f7-a364-aaadb38a13f2", "label": "One Asset, End to End ¶3", "title": "One Asset, End to End — From prompt to candidate", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## From prompt to candidate\n\nThe documented rebuild began with a visual problem. Tidebreak's earlier production pass had become technically elaborate without looking strong enough beside the other games. The revised direction replaced its clockwork guardian and simpler symbols with a regal sea-creature treatment and richer painted jewelry. The retained original environment paintings provided continuity while the symbol and character presentation changed.\n\nThe generative contribution to this wild was its painted design. The retained master is a detailed image with a transparent surround, not a complete animated object. It supplies the dragon's recognizable profile, colors, ornament, and much of the impression of expensive material. The result can be inspected as a strong illustration, but that inspection alone says nothing about whether the symbol will move consistently on the reels.\n\nThe production record identifies the chosen master and the replacement direction. It does not establish a numerical comparison of every discarded wild candidate, so we do not invent a selection tournament or a success rate. What can be said is that this design was retained in the rebuilt set and carried through the subsequent tools. The quality decision was a visual choice, not an automatic consequence of the generator producing a valid image.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "18493443-9aa4-493f-b605-16aa85c506b7", "label": "One Asset, End to End ¶4", "title": "One Asset, End to End — From prompt to candidate", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "The quality decision was a visual choice, not an automatic consequence of the generator producing a valid image.\n\nThere was also a boundary on what the painting needed to solve. It did not have to invent an unseen back of the dragon or provide an anatomically complete model. The intended delivery was a relief emblem. Judging it against that purpose makes more sense than pretending a single attractive view contains every surface required by a full-round creature.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "b6561502-7f2b-4472-adc4-ecf4805522ad", "label": "One Asset, End to End ¶5", "title": "One Asset, End to End — Into the tool", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## Into the tool\n\nThe artwork first became a consistently framed transparent symbol image. This preparation controlled its placement and working area before the rendering stage. Normalization was not a new art direction: it made the chosen image usable as an input without asking generation to rediscover the same composition every time a delivery requirement changed.\n\nBlender then supplied a textured solid relief representation, including a backing and visible depth. The painted face retained its decorative detail. Controlled orientation and moving illumination provided the performance. For this Tidebreak symbol, the effect is rigid jewelry motion rather than a dragon's face changing expression or a full animal moving through space. That is an important distinction between what the image suggests and what the geometry actually represents.\n\nThe scene also established a repeatable relationship among the emblem, camera, and lights. Instead of producing a new interpretation of the dragon for each frame, rendering followed the same scene through its authored motion. This preserved the source design while creating changing highlights and perspective cues. It did not turn the asset into a freely viewable sculpt or guarantee a pleasing result from arbitrary camera positions.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "6bf39da1-1a2f-4e80-a365-45255d81619b", "label": "One Asset, End to End ¶6", "title": "One Asset, End to End — Into the tool", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "It did not turn the asset into a freely viewable sculpt or guarantee a pleasing result from arbitrary camera positions.\n\nThe rendered sequence was packaged as a browser-ready animation atlas, with a settled image drawn from the same opening pose. That pairing matters because an animated symbol eventually has to become a still symbol again. A beautiful moving asset and a different beautiful poster can still make an awkward pair. Keeping the representations related addresses that discontinuity at production time rather than disguising it with a larger flash.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "64934ad8-35c9-4cbf-b489-6d1dcf2c350d", "label": "One Asset, End to End ¶7", "title": "One Asset, End to End — What shipped", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## What shipped\n\nWhat shipped is a settled WebP image and an associated WebP animation atlas, not a Blender process running inside the game. The player downloads presentation media. The local production scene remains useful for revision, but it is not required to play. This separation allows expensive preparation to happen before delivery while leaving runtime responsible for selecting and displaying the appropriate state.\n\nThe wild's motion frames are actual Blender renders rather than optical-flow inventions between generated pictures. That statement is specific to this symbol workflow. Tidebreak's separate character video uses a different production path, so it would be inaccurate to generalize the symbol's provenance to every moving thing on the screen.\n\nThe documented local review covered symbol loading, feature transitions, and multiple viewport sizes as part of the wider rebuild. Those checks support a delivery account, not a claim that the asset has been universally playtested or achieved an objective quality tier. The most useful before-and-after comparison is therefore functional: the master supplies the look, while the delivered pair adds controlled motion, consistent framing, and a defined settled representation that the game can use.", "document": { "id": "cmud5rdnt0004qbiizga21857", "title": "One Asset, End to End", "summary": "- The document outlines the end-to-end workflow for rebuilding Tidebreak’s sea-dragon wild symbol, from initial design prompts and generative artwork through tool-based rendering to the final packaged media that ships with the game.\n- It is a production/engineering case study or conference-style material from Offphase Games, describing real-world asset delivery and public-facing references (publicly served files and assets)." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "34c57296-cd02-4e5d-acb3-9baf8390fc75", "label": "Why the Tool Has to Be Driven, Not Used ¶2", "title": "Why the Tool Has to Be Driven, Not Used — Driving versus using", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## Driving versus using\n\nDriving Blender means describing repeatable scene construction and rendering work so the tool can execute it consistently. Using Blender by hand means making those changes through its interface. Both can produce excellent work. Offphase Games' reason for driving the tool is not that manual artists lack precision; it is that a related set of game assets creates many opportunities for small production differences to become visible.\n\nAs of September 2026, the documented Tidebreak and Sunforge production passes include editable scene construction and saved Blender scenes. These records separate the production representation from the game delivery. We can inspect the scene, change the intended behavior, and produce another delivery without treating the last exported image as the only surviving source.\n\nThe decisions still need direction. Automation does not decide that a symbol's turn feels confident rather than nervous, that a highlight is too bright, or that the character design belongs in this game. It executes choices. A repeated mistake is still a mistake, and making it easy to repeat does not turn it into a style.", "document": { "id": "cmud5rgw80007qbiiqyu06gvv", "title": "Why the Tool Has to Be Driven, Not Used", "summary": "- The document argues that automation should drive the tool (describing repeatable scene construction and rendering) rather than being used manually, and it explains the benefits and limits of this approach for Tidebreak and Sunforge production passes. \n- It is an informational written piece from Offphase Games, dated September 22, 2026, describing production methodology and the costs and benefits of automation in this pipeline." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "4eee65ec-6e20-45f1-8f82-60cbca4ca42c", "label": "Why the Tool Has to Be Driven, Not Used ¶3", "title": "Why the Tool Has to Be Driven, Not Used — What automation buys", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## What automation buys\n\nThe immediate benefit is consistency across a set. A reel symbol is rarely delivered alone. Its neighbors need related framing, visual weight, and motion treatment, while still being distinguishable from one another. A controlled production process can carry common assumptions across the set rather than depending on someone remembering every setting during every export.\n\nIt also makes revision more tractable. If the shared treatment needs adjustment, the team can revise the production representation and regenerate affected outputs. That is different from repairing a collection of finished images independently. It reduces one source of accidental divergence, although each revised asset still needs inspection. A consistent camera can faithfully reveal an inconsistent source design.\n\nThe relationship between the still image and motion is another benefit. Producing both from the same representation gives us a concrete place to manage continuity. Packaging and integration checks can then verify whether the game receives the expected assets. The tool supplies controlled media; the browser implementation remains responsible for using that media at the right moment.\n\nNone of these benefits requires claiming that manual production cannot scale. Skilled teams do it. Our choice is to automate repeatable work in this particular pipeline and reserve review for decisions that cannot be reduced to a file specification. We do not have a controlled labor-cost study that justifies a numerical productivity multiplier for this session.", "document": { "id": "cmud5rgw80007qbiiqyu06gvv", "title": "Why the Tool Has to Be Driven, Not Used", "summary": "- The document argues that automation should drive the tool (describing repeatable scene construction and rendering) rather than being used manually, and it explains the benefits and limits of this approach for Tidebreak and Sunforge production passes. \n- It is an informational written piece from Offphase Games, dated September 22, 2026, describing production methodology and the costs and benefits of automation in this pipeline." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "f5d4a8dc-91e6-4606-8589-13d1f99bbd2f", "label": "Why the Tool Has to Be Driven, Not Used ¶4", "title": "Why the Tool Has to Be Driven, Not Used — What automation costs", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## What automation costs\n\nThe first cost is setup. A repeatable scene-building process takes effort before it produces useful output. It must express the intended camera, materials, motion, and output relationships. For a single disposable illustration, that investment may not be worthwhile. The benefit grows when related assets or repeated revisions make consistent reconstruction valuable.\n\nThe second cost is brittleness. A changed source image, missing dependency, or different tool behavior can invalidate an assumption. Retaining production sources helps diagnosis, but does not eliminate maintenance. Repeatable instructions are also not a promise of identical bytes across every hardware, renderer, and software revision. The delivered output still needs checking.\n\nThe third cost is false confidence. Tidebreak's rejected earlier pass demonstrates that more modeled machinery and a native rig can still produce a weaker visual result. We cannot use automation coverage as a substitute for looking at the game. The best reason to drive a tool is to make a good decision easier to carry through, not to avoid deciding whether it was good.", "document": { "id": "cmud5rgw80007qbiiqyu06gvv", "title": "Why the Tool Has to Be Driven, Not Used", "summary": "- The document argues that automation should drive the tool (describing repeatable scene construction and rendering) rather than being used manually, and it explains the benefits and limits of this approach for Tidebreak and Sunforge production passes. \n- It is an informational written piece from Offphase Games, dated September 22, 2026, describing production methodology and the costs and benefits of automation in this pipeline." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "cb09a002-15be-48a6-8939-6689b02bb5d3", "label": "What the Generative Half Gets Wrong ¶2", "title": "What the Generative Half Gets Wrong — The recurring failures", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## The recurring failures\n\nThe recurring production trap is mistaking an attractive candidate for a finished asset. As of September 2026, Offphase Games' review experience includes symbols that do not read clearly enough in the game, performances that need better continuity, and technically elaborate replacements that look worse than what they replace. Not all of these are model failures. Some are selection, direction, or integration failures, and blaming generation for every weak result hides the useful diagnosis.\n\nOne problem is detail without hierarchy. A large illustration can reward close inspection while becoming an indistinct patch on a reel. Ornament, highlights, and background contrast compete with the identifying shape. More detail does not necessarily mean more recognition. We evaluate the symbol at its working size and beside its neighbors, not just as a large standalone image.\n\nAnother problem is identity across time. Generated motion can make a compelling short performance while introducing changes that would be unacceptable for a stable game symbol. A matching character anchor helps establish the intended subject; it does not prove that every subsequent frame preserves it. That is why our Tidebreak account distinguishes generated character video from symbols animated through controlled relief geometry.", "document": { "id": "cmud5rjro000aqbiiou6jhexx", "title": "What the Generative Half Gets Wrong", "summary": "- It covers recurring failures in generated game assets (and related evaluation and repair approaches), illustrating problems like detail without hierarchy, identity across time, and the idea that complexity can’t fix weak direction. \n- It is an internal review policy/document from Offphase Games, dated September 22, 2026, describing how they evaluate and decide when to regenerate, edit, or replace assets." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "73df3630-1f9c-4650-a293-4a9c43eb01f5", "label": "What the Generative Half Gets Wrong ¶3", "title": "What the Generative Half Gets Wrong — The recurring failures", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "That is why our Tidebreak account distinguishes generated character video from symbols animated through controlled relief geometry.\n\nA third problem is assuming complexity will repair weak direction. The rejected Tidebreak pass had more full-round modeling and a native rig. Its replacement used richer painting and a narrower geometric treatment. This was not a victory of generation over modeling. It was a reminder that the representation has to serve the desired appearance rather than become the reason for approving it.", "document": { "id": "cmud5rjro000aqbiiou6jhexx", "title": "What the Generative Half Gets Wrong", "summary": "- It covers recurring failures in generated game assets (and related evaluation and repair approaches), illustrating problems like detail without hierarchy, identity across time, and the idea that complexity can’t fix weak direction. \n- It is an internal review policy/document from Offphase Games, dated September 22, 2026, describing how they evaluate and decide when to regenerate, edit, or replace assets." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } }, { "id": "b5ea104e-6060-46ec-9d76-3c512d4621fa", "label": "What the Generative Half Gets Wrong ¶4", "title": "What the Generative Half Gets Wrong — Fixed downstream, or regenerated", "source": "Offphase Games, September 22, 2026", "pageNumber": null, "text": "## Fixed downstream, or regenerated\n\nWe repair a delivery problem downstream when the underlying design is sound and the required change is explicit. Framing, output dimensions, file format, or the relationship between a settled image and an animation can be production problems rather than reasons to invent a different creature. Controlled tools are useful precisely because they can change those properties without reopening the entire visual decision.\n\nWe reconsider the candidate when the defect is in its identity or composition. If the silhouette is weak, the object is hard to recognize, or the design does not fit the game, a clean export will preserve the wrong thing more efficiently. Regeneration is one option; editing or a different artistic approach can also be appropriate. There is no rule that every rejected asset deserves another model call.\n\nFor unstable motion, the choice depends on what must remain fixed. An expressive character performance may justify another generated take. A rigid emblem may be better served by a controlled representation. Neither choice excuses checking the transition back into gameplay. We accept the result only when the asset and its use work together. This is a review policy, not a measured claim about model failure rates or a guarantee that every published asset is beyond improvement.", "document": { "id": "cmud5rjro000aqbiiou6jhexx", "title": "What the Generative Half Gets Wrong", "summary": "- It covers recurring failures in generated game assets (and related evaluation and repair approaches), illustrating problems like detail without hierarchy, identity across time, and the idea that complexity can’t fix weak direction. \n- It is an internal review policy/document from Offphase Games, dated September 22, 2026, describing how they evaluate and decide when to regenerate, edit, or replace assets." }, "knowledgeBank": { "slug": "off-phase-games", "name": "Offphase Games" } } ] }