Image tools: work around OWUI 0.9.x files-event regression

Open WebUI 0.9.0 introduced chat-history virtualization that
unmounts off-screen assistant messages and reconstructs them from
persisted shape; `files` attached mid-stream by a tool don't
survive the round-trip — the image flashes in during streaming
and disappears the moment the message commits.

Both image tools now upload via Open WebUI's file store as before
but surface the result as a markdown image injected into the
assistant message via a `message` event, which is part of the
persisted shape and renders reliably across remounts. Trade-off:
loses the file-attachment chrome (thumbnail + download button).
Each tool has a TODO marking the swap site with the original
`files` payload inlined for one-line revert once upstream fixes
the regression.

smart_image_gen.py 0.7.8 -> 0.7.9
smart_image_pipe.py 0.1.0 -> 0.1.1

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-04-26 06:03:21 -05:00
parent 28f370a80b
commit d034700af9
2 changed files with 51 additions and 15 deletions

View File

@@ -1,7 +1,7 @@
""" """
title: Smart Image Generator & Editor (ComfyUI) title: Smart Image Generator & Editor (ComfyUI)
author: ai-stack author: ai-stack
version: 0.7.8 version: 0.7.9
description: Generate or edit images via ComfyUI with automatic SDXL description: Generate or edit images via ComfyUI with automatic SDXL
checkpoint routing. Two methods — generate_image (txt2img) and checkpoint routing. Two methods — generate_image (txt2img) and
edit_image (img2img on the user's most recently attached image). The edit_image (img2img on the user's most recently attached image). The
@@ -9,8 +9,8 @@ description: Generate or edit images via ComfyUI with automatic SDXL
score-tag, NoobAI/Illustrious furry, etc. — and each style ships score-tag, NoobAI/Illustrious furry, etc. — and each style ships
with the creator-recommended sampler, scheduler, CFG, steps, CLIP with the creator-recommended sampler, scheduler, CFG, steps, CLIP
skip, prompt-prefix dialect, and negatives. The image is uploaded skip, prompt-prefix dialect, and negatives. The image is uploaded
to Open WebUI's file store and surfaced via a `files` event (the to Open WebUI's file store and surfaced as a markdown image
canonical pattern used by Open WebUI's own image-gen path); the appended to the assistant message via a `message` event; the
function return is a short confirmation so the LLM doesn't try to function return is a short confirmation so the LLM doesn't try to
describe or re-emit the image. describe or re-emit the image.
required_open_webui_version: 0.5.0 required_open_webui_version: 0.5.0
@@ -586,14 +586,22 @@ async def _push_image_to_chat(
event_emitter: Optional[Callable[[dict], Awaitable[None]]], event_emitter: Optional[Callable[[dict], Awaitable[None]]],
) -> bool: ) -> bool:
""" """
Surface a generated image in the chat using Open WebUI's canonical Surface a generated image in the chat: upload the bytes via the
pattern: upload the bytes via the internal file store, then emit a internal file store, then inject a markdown image referencing the
`files` event referencing the served URL. This is the same path Open served URL into the assistant message via a `message` event.
WebUI's own image-generation code uses (utils/middleware.py ~1325).
Returns True if the image was uploaded and emitted via the files We deliberately don't use the `files` event (Open WebUI's own
event. Returns False if anything is missing — caller should fall image-gen path). Open WebUI 0.9.x added chat-history virtualization
back to a data-URI markdown message in that case. that unmounts off-screen messages and reconstructs them from
persisted shape — and `files` attached to an assistant message
mid-stream by a tool don't survive that round-trip. The image
flashes in during streaming and disappears the moment the message
commits. Markdown in `message.content` is part of the persisted
shape, so it renders reliably on every remount.
Returns True if the image was uploaded and emitted. Returns False
if anything is missing — caller falls back to a data-URI markdown
message (same `message` event path, just inline bytes).
""" """
if not (_OPENWEBUI_RUNTIME and request and user_dict and event_emitter): if not (_OPENWEBUI_RUNTIME and request and user_dict and event_emitter):
return False return False
@@ -630,9 +638,23 @@ async def _push_image_to_chat(
"get_file_content_by_id", id=file_item.id "get_file_content_by_id", id=file_item.id
) )
# TODO(open-webui virtualization fix): once chat-history
# virtualization survives `files` events from tool calls again,
# swap back to the `files` payload below. The `files` event
# gives proper file-attachment chrome (thumbnail card + download
# button) that the markdown-image path lacks. Verify the fix by
# emitting `files`, then refreshing the page AND scrolling the
# message off-screen and back — both must keep the image
# visible. Track upstream: github.com/open-webui/open-webui
# release notes mentioning tool calls, `files` events, or
# virtualization. To restore, replace the block below with:
# await event_emitter({
# "type": "files",
# "data": {"files": [{"type": "image", "url": url}]},
# })
await event_emitter({ await event_emitter({
"type": "files", "type": "message",
"data": {"files": [{"type": "image", "url": url}]}, "data": {"content": f"\n![{filename_prefix}]({url})\n"},
}) })
return True return True
except Exception: except Exception:

View File

@@ -1,7 +1,7 @@
""" """
title: Smart Image Studio (Pipe) title: Smart Image Studio (Pipe)
author: ai-stack author: ai-stack
version: 0.1.0 version: 0.1.1
description: Deterministic image-gen / edit / inpaint pipe — no LLM in the description: Deterministic image-gen / edit / inpaint pipe — no LLM in the
loop for the routing decision. Registers as a model in the chat-model loop for the routing decision. Registers as a model in the chat-model
dropdown ('Image Studio (Pipe)'). Reads the user's message + attached dropdown ('Image Studio (Pipe)'). Reads the user's message + attached
@@ -420,9 +420,23 @@ async def _push_image_to_chat(raw, prefix, request, user_dict, metadata, event_e
) )
file_item = await result if inspect.iscoroutine(result) else result file_item = await result if inspect.iscoroutine(result) else result
url = request.app.url_path_for("get_file_content_by_id", id=file_item.id) url = request.app.url_path_for("get_file_content_by_id", id=file_item.id)
# TODO(open-webui virtualization fix): once chat-history
# virtualization survives `files` events from tool calls again,
# swap back to the `files` payload below. The `files` event
# gives proper file-attachment chrome (thumbnail card + download
# button) that the markdown-image path lacks. Verify the fix by
# emitting `files`, then refreshing the page AND scrolling the
# message off-screen and back — both must keep the image
# visible. Track upstream: github.com/open-webui/open-webui
# release notes mentioning tool calls, `files` events, or
# virtualization. To restore, replace the block below with:
# await event_emitter({
# "type": "files",
# "data": {"files": [{"type": "image", "url": url}]},
# })
await event_emitter({ await event_emitter({
"type": "files", "type": "message",
"data": {"files": [{"type": "image", "url": url}]}, "data": {"content": f"\n![{prefix}]({url})\n"},
}) })
return True return True
except Exception: except Exception: