Repository navigation
Conversation
The two VL nodes and the Tile Test: Captions node take the image's prompt and write it into a caption preset's instructions through captions.with_prompt. The prompt reaches the VL model's question only, never the DiT. A blank prompt against a preset that asks for one fails before any GPU work, and the caption pass refuses an unfilled placeholder. The VL nodes now resolve the preset themselves and hand it to refine_image, so the engine's own resolve is the direct-caller path. settings.toml ships a new first preset, prompted, with the owner's wording under test. standard moves to second and the shipped Krea 2 refine workflow names it. Tile Test: Captions gains tiles in and out, the custom instructions option, one progress bar through build_caption_ledger, and keeps the style caption apart from the tile captions in its listing.
The VL nodes and the Captions node read prompt from an optional socket, and the vlm_method selector now labels the first preset like the rest. The README documents both and lists the Tile Test nodes under caption presets instead of in their own section.
…test node and a caption timing harness
…dd the tile tags benchmark
…ompt-only tags at 0.9999
…est workflow's renamed sockets
…checks and status text, reject unsupported samplers at queue time, clean up the Markdown outputs, share duplicated engine code, update docs
…nd progress on how the run ended, move the bench to fp8 and the shared bootstrap, and add output tooltips and search aliases
…o fallback, measured on the fp8 encoder, and show the translation and the fallback in the Captions node's tag outputs
…n arm with no units shared with the base instead of dividing by zero
…it no longer runs
…s templates, and point the README at them
…ds as one pair, and show the thing check as p(thing)
Owner
Author
|
Major update v1.8.0 that includes the updated VL method that samples the tile and full image, TEST nodes added, settings file with custom prompts and tile settings for VL nodes, adds style prompt across tiles, pulls prompts from your positive prompt, I created a https://pypi.org/project/logit-classifier/ that I use to tell what parts of the main prompt need to be pulled into the current tile and uses a fast logit Noul that is inspired by Jev but works with the Qwen3 4b VLM that is already being used. Captions are now faster and far more accurate, along with other optimizations. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.