Image Studio: default base model → huihui_ai/qwen3-vl-abliterated:8b

User confirmed this model works end-to-end after the multi-base-model
search. Settled on it because Qwen 3 VL's fine-tune lineage isn't
damaged by abliteration the way Qwen 3.5's is, so it both calls tools
reliably AND won't refuse to dispatch on NSFW edit prompts.

Updated:
  - image_studio.json base_model_id → huihui_ai/qwen3-vl-abliterated:8b
  - init-models.sh: pulls the abliterated VL model in place of the
    non-working standard qwen3.5:9b
  - image_studio.md: setup table base-model row + vision-section
    'why this and not the alternatives' explanation

function_calling stays default and tool_choice required. Operator
can flip to native + drop tool_choice once they've verified the new
base behaves with structured tool calls (which would also remove the
need for a separate Task Model for title generation).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
57_WolveandClaude Opus 4.7 committed 2026-04-19 17:56:16 -05:00
1 parent ec6108888a
commit f5a222fe6f
3 files changed
+25 -17

No files matched your search

@@ -38,7 +38,7 @@ prompts. Verify after import:
| Field | Value |
| ----- | ----- |
| Name | `Image Studio` |
| Base Model | `qwen3.5:9b` (6.6 GB, vision + native tool calling). Pull via `init-models.sh` first. The abliterated variant of the same model breaks tool-call format adherence; this is the standard build. |
| Base Model | `huihui_ai/qwen3-vl-abliterated:8b` (Qwen 3 VL base, abliterated, vision + tools). Pull via `init-models.sh` first. The Qwen 3 VL fine-tune lineage isn't damaged by abliteration the way Qwen 3.5 is, so it both calls tools reliably AND won't refuse to dispatch on NSFW edit prompts. |
| Description | `Image generation and routing across SDXL checkpoints.` |
| System Prompt | Paste the block from [System prompt](#system-prompt) below. |
| Tools | enable **only** `smart_image_gen` |
@@ -178,20 +178,26 @@ suggestions return, and autocomplete lights up.
The shipped preset sets `meta.capabilities.vision: true` so Open WebUI
allows users to attach images to chats with this model. Two paths:
### Default — `qwen3.5:9b`
### Default — `huihui_ai/qwen3-vl-abliterated:8b`
The shipped preset uses standard (non-abliterated) Qwen 3.5 9B as
the base — vision-capable, 256K context, native tool calling
working. Preseed via `init-models.sh`.
The shipped preset uses huihui_ai's abliteration of Qwen 3 VL as
the base — 8B params, vision-capable, native tool calling working,
and won't refuse to dispatch the tool when the user's edit prompt
is NSFW. Preseed via `init-models.sh`.
**Why not the abliterated 9B?** The abliteration process mangles the
**Why not the Qwen 3.5 abliterated 9B (huihui_ai/qwen3.5-abliterated:9b)?**
Same maintainer, but the abliteration on Qwen 3.5 mangles the
function-call template, causing the model to either refuse to call
tools or emit malformed `<function=...>` XML that Open WebUI's parser
can't recognise. Image generation doesn't actually need an
uncensored LLM — the SDXL checkpoints in ComfyUI are what produce
image content, and they're already uncensored. The LLM's only role
here is to call the right tool with the right args; the standard
build does that more reliably.
tools or emit malformed `<function=...>` XML that Open WebUI's
parser can't recognise. The Qwen 3 VL fine-tune lineage is
different and doesn't take that damage from abliteration.
**Why not standard `qwen3.5:9b`?** The standard (non-abliterated)
Qwen 3.5 calls tools reliably but its safety training refuses on
many image edit prompts even though the LLM's only job is dispatch
(the actual image content is generated by the SDXL checkpoint, which
the LLM never sees). Abliterated VL gets us both reliable tool
calling AND a cooperative dispatcher.
**Qwen 3.x quirk:** thinking mode is on by default. The shipped
system prompt starts with `/no_think` to suppress it. If the model