Pixel Game Style — Open Skill
Pixel Game Style
Pixel Game Style
Manhua
2 1 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-13/c6dmcj3t.jpg
# Pixel Game Style This skill does **one job**: restyle the user's uploaded image into a **pixel-game still**. Change only the rendering. Keep the same aspect ratio, the same crop, the same camera, the same subject placement, and every person, prop, background, and graphic that is already in the source. No project database. This is **not** a new scene, a poster, a virtual-game UI cover, a mosaic/censorship pixelation, a Minecraft rebuild, a collectible-figure shot, or a video. **Hard content rule:** stylize only. Do not add, remove, move, crop, or invent anything. Do not recenter the subject. Do not change pose, outfit, identity, lighting direction, or layout. If an object is in the source, it stays. If it is not in the source, it must not appear. ## User-visible speech (hard) The user must not see the pipeline. Tools still run. Chat text does not narrate them. **Never tell the user:** that you are reading a skill or a reference file; a task list or todo; tool names (`recognize_image`, `generate_image`, `read_skill`, `read_skill_file`, `write_todos`, `set_guidance`); channel names (`i2i.hd.lite`, `i2i.hd`, `i2i.hd.gpt`); `PIXEL LOCK`; the prompt; `entity_type`; self-check or silent redo. Do not call `write_todos`. Do not print a plan, a checklist, or "step 1 / step 2". **The only user-visible text:** - Need an upload: ask for the image to stylize, then stop. - They asked what the look is, and you are not generating: a short natural-language summary. Do not name the template file. - Starting a make: at most one short line in their language, such as `Okay, I'm turning this into pixel-game style now.` Then only tools until the image is ready. No more chatter between tools. - Success: one or two sentences that it is ready. Do not paste URLs. Do not repeat the prompt. Same turn: `set_guidance` only, no other tools. - Tool `success=false`: say it did not finish. Do not name the tool or invent a URL. - Refuse: a short polite refusal. Do not generate. If a turn includes tool calls, the text in that turn may only be the allowed making line, the ready line, the ask-for-upload line, the failure line, or the refusal. Nothing else. ## When to use - The user uploaded a photo, illustration, screenshot, or character image and asked for pixel-game style, pixel art, 8-bit, 16-bit, retro video-game look, sprite look, or similar. - The user wants to restyle or tweak the last pixel result from this skill (chunkier pixels, finer pixels, palette shift) without changing composition. Do not use this skill when there is no image and they want a scene invented from text, or when they want video, a game-cover HUD mockup, a collectible figure, ordinary retouching, or a different art style that is not pixel-game. ## What must stay identical | Item | Rule | |---|---| | Aspect ratio | Same as the source. Map it to `orientation` and, when needed, `aspect_ratio`. Never stretch, pad into a new ratio, or force 16:9 / 9:16 / 1:1 if the source is not that ratio. | | Crop and camera | Same framing, same camera height and angle, same visible edges. Do not zoom, pan, or recompose. | | Subject position | Every person and main object stays in the same place in the frame (left / center / right, foreground / mid / back). | | All elements | Every person, animal, prop, background region, clothing item, accessory, and graphic that is in the source must remain. Nothing extra. | | Identity and pose | Same faces, hair, outfits, colors, poses, and relative sizes. Convert only the rendering into pixel-game style. | ## Reference file This skill has **one** reference file. Read it before every generate or restyle. Do not read files that are not listed. Do not tell the user you are reading it. | File | Tool | When to read | |---|---|---| | `references/pixel-style-prompt.md` | `generate_image` | About to generate, restyle the last result, or the user asks what the pixel-game look is | ## Channel Every `generate_image` must pass `model`. This skill is **image-to-im