PromptForge is out: free, local, and it writes for each model
Every model wants its prompt written differently: booru tags for Illustrious, prose for Krea2, timed shots for MiniMax H3, one flowing paragraph for LTX. Most prompt helpers write the same thing for all of them.
PromptForge writes for the model you picked. You give it a short brief, choose the model, and get a prompt in that model’s own format. It runs on your own machine with your own local LLM.
Used Darksidewalker’s MMH3 Director? Then you’ve already used a piece of PromptForge. We built the H3 Forge in that node, and it runs the same H3 prompt writing. PromptForge is the full version, for every model.
Our own H3 prompt model: PromptForge-H3-4B. We also trained a small 4B model just for writing MiniMax H3 prompts. It’s fast, it can see your reference pictures, and it runs on modest GPUs, in PromptForge or the H3 Forge.
👉 https://civitai.com/models/2997688/promptforge-h3-4b
Models: Illustrious XL, Anima, Krea2, MiniMax H3, Wan 2.2, LTX 2.3
Also does: character looks from your brief, reference pictures, rendering in ComfyUI, picture editing, tag search, chat, and LoRA training through TrainFlow. You can host it for your friends.
🐳 Docker Hub: https://hub.docker.com/r/trashai/promptforge
💻 GitHub: https://github.com/yoimtrashgaming-prog/PromptForge-Public
Setup instructions are on GitHub. Big thanks to azmagane and Darksidewalker ❤️
Free and open source (GPL-3.0).
Feedback wanted. These have had the least testing so far, so this is where your feedback helps most:
LTX 2.3: whether the paragraphs come out right, and how they render
Wan 2.2: how well the prompts hold up across different scenes
Anima: the mix of tags and natural language, and whether characters come out right
PromptForge-H3-4B: how its H3 prompts compare to a bigger model’s