57_WolveandClaude Opus 4.7
ec6108888a
Vision-capable LLMs misclassify rendered subjects when picking a style — observed: model called juggernaut for an edit on a furry-il generation because the rendered character looked 'photoreal-ish' to its vision encoder. Each visual judgment is independent so styles flip mid-chat. Flipped resolution order in edit_image so inheritance from the prior generate_image / edit_image call DOMINATES the LLM's explicit style arg. The LLM's choice only wins when there's nothing to inherit (first edit in a chat, fresh user upload). Workaround for legitimate style changes is starting a new chat. System prompt updated to match: tells the LLM that style inheritance is enforced, that passing style on follow-up calls is ignored, and that user requests for style change require a new chat. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
comfyui-nvidia
ComfyUI image-generation backend, NVIDIA-accelerated, fronted by Open WebUI for multi-user chat and image generation/editing.
Built from the official ComfyUI manual install for
NVIDIA — no
third-party base image. CI publishes the image to
git.anomalous.dev/alphacentri/comfyui-nvidia on every v* tag (see
.gitea/workflows/release.yml).
Repository layout
| Path | What |
|---|---|
Dockerfile |
ComfyUI on NVIDIA, manual-install pattern |
workflows/ |
txt2img + img2img workflow JSONs and node mappings |
deployments/ai-stack/ |
The deployment — compose, Caddyfile, env, model preseed |
.gitea/workflows/ |
Release pipeline (build & push image on tag) |
Deploy
The full stack — Caddy + Ollama + ComfyUI + Open WebUI (+ optional
Anubis) — lives under deployments/ai-stack/.
Bring-up steps, host prerequisites, Open WebUI workflow wiring, and
gotchas are in deployments/ai-stack/README.md.
Replaces
This repo supersedes the previous figment + segment + Forge stack. ComfyUI's node graph covers everything those services provided (txt2img, img2img, inpaint, mask generation via SAM/GroundingDINO custom nodes), and Open WebUI talks to it natively.