Film Production Studio
Art Studio
Every picture the studio makes, from a 2D drawing to a 3D game asset, directed by one lane and drawn by open software on our own machines. Direction is written down, so the look holds across every session and every picture.
Where it stands
Each piece run by run: the strongest picture of each run and the director's verdict. Open a run to see every candidate, the numbers, what must change, and the exact words.
Lighthouse Workshop · painted-storybook



How a picture is made
| 1. The bible | Each look is written once: its feel, medium, shapes, palette by role, light, composition, do and don't, and the exact style words. One look, one account. |
|---|---|
| 2. The piece | What is in the picture, where the eye goes, the shot and the beat. Never a typed prompt. |
| 3. Compile | Piece and bible become the model's words, in a fixed order, the same every time. |
| 4. Draw | Several candidates on an open model, each saved with everything needed to make it again. |
| 5. Judge | The machine measures palette distance, value grouping, focal contrast and likeness to the touchstones. |
| 6. Direct | The Art Director looks at every candidate whole, passes one or sends the piece back with what must change. The verdict goes in the bible's log for good; a pass can become a touchstone. |
Looks
Painted Storybookstudio · gouache and watercolour on cold-press paper · warm, safe, curious; a picture a child would want to step intoModels
| Model | Job | Licence | Use | Runs on |
|---|---|---|---|---|
| z-image-turbo | photoreal and illustration, text in the picture; 8 steps | Apache-2.0 | yes | card, par1-cpu (slow: bf16 is emulated on par1's Zen 3) |
| flux2-klein-4b | text to image and editing with references in one model; 4 steps | Apache-2.0 | yes | card, par1-cpu |
| hidream-o1 | photoreal and editing, the best MIT model that fits the L4 (arena 979) | MIT | yes | card |
| chroma1-hd | painted and illustrated styles, real negative prompts; base for our own style LoRAs | Apache-2.0 | yes | card |
| kromoss | broad-style fine-tune of MOSS-Image, 8 steps with its DMD2 LoRA | Apache-2.0 | yes | card |
| illustrious-xl-2 | anime and cartoon, the deepest character knowledge; LoRA per character; lineart and pose control | CreativeML OpenRAIL-M | yes | par1-cpu, card |
| noobai-xl | anime with the widest character and artist knowledge | Fair AI Public Licence 1.0-SD (fine-tunes we train stay private, never served) | yes | par1-cpu, card |
| qwen-image-2512 | text and lettering in the picture, layouts, posters | Apache-2.0 | yes | card (fp8 or nunchaku 4-bit) |
| qwen-image-edit-2511 | editing with up to three references: a character on model across poses, outfits and scenes | Apache-2.0 | yes | card (nunchaku 4-bit) |
| qwen-image-layered | splits a finished picture into editable RGBA layers: cut-out parts for the rigs | Apache-2.0 | yes | card |
| glm-image | accurate lettering with a glyph encoder | MIT | yes | card |
| flux-3-klein | the highest-ranked open model we may use (arena 1041); editing with up to 10 references | FLUX Kommunity Licence: free under US$5M revenue | cap | card |
| krea-2-turbo | aesthetic illustration, ships watercolour and kids-drawing style LoRAs | Krea 2 Community Licence: free under US$1M revenue, content filter required | cap | card |
| birefnet | cuts a subject from its background (sprites, cut-out parts, compositing) | MIT | yes | par1-cpu, card |
| seedvr2 | restoration and upscaling | Apache-2.0 | yes | card |
| sam-3.1 | masks by text or click | SAM Licence (commercial allowed; no military or ITAR use) | yes | card |
| omnisvg-1.1 | text or picture to editable SVG | Apache-2.0 | yes | card |
| trellis-2 | picture to textured PBR 3D mesh, game and film assets | MIT | yes | card (L40S; 24 GB is its floor) |
| pixal3d | picture-faithful 3D mesh whose front matches the art | MIT (not a Hunyuan licence) | yes | card |
| minimax-h3-image | the top open model on both arenas | MiniMax H3 Community Licence: excludes the US, EU, UK and Korea | never | |
| qwen-image-2.1 | text to image, Sept 2026 | Qwen Research Licence (non-commercial) | study | |
| anima | the best small anime model | non-commercial | study | |
| pony-v7 | anime and cartoon | excludes professional video production | study | |
| insightface-identity-adapters | PuLID, InstantID, InfiniteYou, IP-Adapter FaceID | InsightFace face models are non-commercial | study | |
| microsoft-lens | text to image | uses OpenAI gpt-oss-20b as its text encoder | never |
Tools
| Tool | What | Licence |
|---|---|---|
| comfyui | the node runtime every open image workflow is written for, driven headless over HTTP | GPL-3.0 |
| diffusers | Python pipelines for scripted jobs | Apache-2.0 |
| imagemagick | every raster operation | ImageMagick |
| libvips | fast large-image processing | LGPL-2.1 |
| vtracer | colour raster to vector SVG | MIT |
| potrace | line art to clean vector curves | GPL-2.0 |
| blender | 3D: models, sets, light, Grease Pencil 2D, toon and line-art rendering | GPL-2.0+ |
| siglip2 | the judge's eye for words and likeness | Apache-2.0 |