Workflows, presets and tests: Skool https://www.skool.com/artfat-ai-6792
· Discord https://discord.gg/WWSe9zSSQ
· YouTube https://www.youtube.com/@ArtfatAi
· X https://x.com/ArtfatAi
You draw a box where something should be, write what it is, and run. That's the whole idea.
A neon sign that has to say exactly "NOODLE BAR". Two friends on the left and right. A bowl of ramen at the bottom. Each one gets its own box, and MOSS puts them where you drew them. Short words in a text box usually come out spelled the way you typed them.
Then you can keep going on the same picture. Switch Edit Source to edit last result, draw a box over the sign, type "ARTFAT", run. The sign changes and the rest of the picture stays as it was, pixel for pixel. Sunglasses on her face, then a straw hat on top, one box at a time. It works the same way on your own photo.
What's inside
- Box canvas with object boxes and text boxes. One button switches between bbox generation and plain generation when you don't need boxes.
- Edit Source with three modes: new image, my photo, edit last result.
- Artfat LLM Prompter v2 with 50+ MOSS presets (phone snapshots, film, editorial, noir, Y2K and more). It writes the full prompt from your boxes and a short idea. You can also turn it off and type your own prompt.
- 6 to 8 steps with a speed LoRA instead of the usual 50.
- SeedVR2 for fine detail and DLSS5 Enhancer as a last polish (NVIDIA RTX 30/40/50). Both start switched off, so the first run is fast and works on more cards. Turn them on with Ctrl+B.
- Notes inside the workflow for every step: settings, nodes, models with download links, and what each switch does.
- Interactive guide in the zip (GUIDE.html): every node with screenshots, every switch explained, step-by-step examples.
First run
- The workflow opens on edit last result, and there is no last result yet. Switch Edit Source to new image for your first picture.
- Editing your own photo? Load Image starts switched off too. Select it, press Ctrl+B, pick your photo and set Edit Source to my photo.
You need
- ComfyUI 0.42 or newer, with Nodes 2.0 turned on (the box canvas is a Nodes 2.0 widget).
- An NVIDIA card. The recommended int8 model needs torch for CUDA 13 (cu130). There are GGUF options for 8 to 12 GB cards in the notes.
- Missing nodes: ComfyUI-Manager → Install Missing Custom Nodes → Install All. A pack is not in that list? Search Artfat in the Manager. Still nothing? Install via Git URL with the links in the workflow notes.
- Missing models: ComfyUI shows a Download button for each one.
Works on Windows and Linux.
Tips
- In edit mode, describe the picture after the change. A prompt that still names the old thing pulls it back into the box.
- Short words work best in text boxes. Give the box some room.
- Keep hands and faces either fully inside a box or fully outside it.
Made by Artfat AI. If you make something with it, post it under this workflow, I'd love to see it.