[bug] Trying to generate a picture in ai-studio with the Ideogram 4 lane results in Image generation failed: HTTP Error 400: Bad Request error

[stage-chuk]

What happened?

Using AI Studio to generate an image using Ideogram 4 results in a "Image generation failed: HTTP Error 400: Bad Request" message. This also happens with several other models/lanes, e.g. Krea 2, Ace-Step.

Steps to reproduce

* Launch AI Studio
* Use an arbitrary lane to generate something

Rig report — paste contents of bash scripts/report.sh [--verify | --soak | --bench | --studio | --full]

Too long to paste here so attached as separate file

Anything else?

rig_report_studio.md

[noonghunna]

Same root cause as #510 — the ComfyUI container was mounting an empty models tree (COMFYUI_ROOT wasn't reaching the sudo'd docker compose, so it fell back to the /mnt default while your models live under the MODEL_DIR-derived path). That's why every lane returned Value not in list: … not in [] → HTTP 400. Fixed in #531 (now on master). Pick it up with git pull && gpu-mode ai-studio — no re-download needed. Full diagnosis + a manual fallback are in #510; I'll close both once you confirm the lanes render. 🙏

[noonghunna]

Both fixes for this are on master now:

• fix(studio): pin COMFYUI_ROOT to .env so ComfyUI mounts the models tree (#510, #530) #531 — the COMFYUI_ROOT models-mount fix that resolves the Value not in list: … not in [] → HTTP 400 you hit here (ComfyUI's model dropdowns were coming up empty).
• fix(studio): derive + pin COMFYUI_OUTPUT_DIR from COMFYUI_ROOT (#510 render-mount) #534 — a follow-on I found while fixing [bug] Trying to generate a picture in ai-studio with the Hi-Dream lane results in Image generation failed: ComfyUI generation error #510: once the models mount correctly, the output side had the same class of bug (COMFYUI_OUTPUT_DIR → the :8189 gallery wasn't derived from COMFYUI_ROOT), so renders would 404 in the gallery. Now derived + pinned too — and it self-heals even if you already picked up fix(studio): pin COMFYUI_ROOT to .env so ComfyUI mounts the models tree (#510, #530) #531.

So a single pull gets the full lane working end-to-end:

git pull
gpu-mode ai-studio

Closing as fixed — please reopen (or ping on #510) if any lane is still off after the pull. Thanks again for the report + the --studio output; it made both diagnoses quick.