CXH_Ic_lora_Joy_batch

CXH_Ic_lora_Joy_batch is the JoyCaption-powered version of the pack's IC-LoRA captioning idea. Where CXH_IC_Lora_Florence2Run does the same job with Florence 2, this one runs the pair through JoyCaption Alpha Two - which produces the higher-quality natural-language captions that matter most for Flux-class training data. If you're building a clothing/garment IC-LoRA dataset and want the best captions this pack can produce, this is the node.

"IC" here is the image-conditioned LoRA convention: every sample is two images - the product shot and the worn shot - stitched side by side, captioned as one frame so the model learns the before/after relationship. The default template spells it out: Realistic style, [cloth-on], the image pair highlights a transformation from a clothing sample photo to the effect of actually wearing it. [image1] {caption} [image2] a female model is wearing the cloth from [image1] with {caption}.

How it works

For each image in dir1 it finds the same-named file in dir2, resizes the second to match the first (when match_image_size is on), stitches the pair together in the chosen direction (right/down/left/up), and runs the Alpha Two pipeline - the same SigLIP → image adapter → Llama generation as the single-image nodes. The caption is inserted into template at {caption}, alongside {text1} and {text2} if you set them, and written as a .txt beside a saved copy of the composite pair in saveDir.

Inputs

• JoyPipeline_alpha - from Joy_caption_alpha_load (which needs the Alpha Two checkpoint in models/Joy_caption_alpha/).
• prompt - the captioning instruction (wire in a Joy_caption_alpha_prompt for styled output).
• dir1, dir2, saveDir - paired source folders and output.
• format (png/jpg), max_new_tokens (default 1024), text1, text2, template, direction, match_image_size.

There are no outputs - this is an output node; results are on disk.

Installing

Standard pack install - ComfyUI Manager (search "Comfyui_CXH_joy_caption") or:

cd ComfyUI/custom_nodes
git clone https://github.com/StartHua/Comfyui_CXH_joy_caption
cd Comfyui_CXH_joy_caption
python -m pip install -r requirements.txt

Restart, then get the Alpha Two checkpoint in place (see Joy_caption_alpha_load - that's the manual download step).

Gotchas

• Filenames must match across folders - missing pairs are skipped with a console print, so verify your counts after a run.
• No cache toggle here. Unlike the standalone batch nodes, this one keeps the Alpha Two pipeline resident for the whole batch - great for speed, but on an 8 GB card it can collide with your checkpoint. Consider running it as a dedicated caption pass.
• match_image_size off yields ragged stitches and muddier captions. Keep it on.
• Reverse-direction captions: run the outputs through CXH_IC_lora_reversal afterward to build the mirrored training set.
• NVIDIA only, as with the whole JoyCaption path.