ComfyUI-ialhabbal Introduction
ComfyUI-ialhabbal is a collection of eight groups of custom nodes for ComfyUI. They help you review prompts, inspect images and video with multimodal models, compare results, extract information from existing images, protect foreground objects during face swaps, process batches, finish portraits, and organize saved artwork.
You can use the nodes separately. For example, you might use Prompt Verify to approve an AI-written prompt before generation, PhotoLab to give the resulting portrait a film look, and Save_It to compare it with the original before saving. You do not need to build a workflow containing the entire suite.
The eight groups are Prompt Verify, Compare, Meta Prompt Extractor, Occlusion Mask, Loader for Batch Image Processing, PhotoLab, Save_It, and ialhabbal_VLLM.
How ComfyUI-ialhabbal Works
ComfyUI workflows pass information between connected nodes. Think of each node as a station in a creative process: one station receives an image or text, performs a task, and passes its result to the next.
- Before generation: Prompt Verify lets you inspect or edit a prompt before the workflow continues. The VLLM nodes can help create or refine text and work with image or video inputs.
- During image preparation: Meta Prompt Extractor can recover prompts and image details from supported files. The batch loader gathers multiple images for processing.
- During editing: Occlusion Mask marks foreground objects that should remain visible in a face-swap workflow. PhotoLab applies photographic and skin-detail effects.
- After generation: Compare helps you inspect two images, while Save_It lets you review and organize the results you keep.
Outputs have specific purposes. A STRING output is text you can connect to another text-using node. An IMAGE output carries image data. A MASK identifies an area for another node to use. CONDITIONING is a prompt encoded for a connected CLIP-based generation workflow. Connect an output to an input of the matching type.
ComfyUI-ialhabbal Features
Prompt Verify: Review a Prompt Before It Runs
Prompt Verify gives you a pause point in your workflow. When text arrives through its text or llm_input connection, the node displays it in an editor and waits for you to approve or change it. Click Submit or press Shift+Enter to continue. This is useful when a wildcard system or language model produces a different prompt on each run and you want final editorial control.
With neither input mode connected, text already in the editor is submitted automatically. If the editor is empty, the node waits for you to enter a prompt. A timeout can submit the current editor text if you do not respond; its default is 60 seconds and its maximum is 2,400 seconds.
The node provides the approved text as a STRING. If you connect an optional CLIP model, it also provides encoded CONDITIONING for a compatible generation workflow. Its prompt library lets you save prompts by category and name, then search, load, rename, delete, export, or import them. The library persists between sessions.
Prompt Verify node and editor (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/prompt_verify_the_node.png) Simple Prompt Verify workflow (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/prompt_verify_simple_workflow.png)
Compare: Inspect Two Images
Connect two images to image_a and image_b to view them side by side. Click an image to open an overlaid comparison, use the 1/2 toggle to switch between versions, and click x to close that view. This makes it easier to judge small differences, such as whether retouching preserved facial detail or a new generation improved the composition.
Meta Prompt Extractor: Revisit Existing Images
Meta Prompt Extractor reads supported PNG, JPG, JPEG, WebP, and JSON files and looks for generation information. It can output a positive prompt, a negative prompt, the image, a mask, its path, its pixel width and height, and a resolution string such as 512x512. This is useful when you want to continue work from an earlier image but no longer have its original workflow open.
Its file browser offers thumbnails, search, sorting, folder favorites, multi-selection, and a metadata preview. You can select a folder with batch_folder or supply exact paths through batch_file_list. The browser also includes file-management actions, drag-and-drop, and a Mask Editor for painting an image mask. A matching <image>_mask.png can provide a mask as well.
The node traces workflow connections when looking for a prompt and uses a fallback scan when needed. An extracted prompt depends on what the source file contains: a file without usable metadata may not yield the text you expect. If you enable use_conditioning, connected positive and negative conditioning take precedence over prompts extracted from the file.
Meta Prompt Extractor file browser (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/mpe_Browse_Files_floating_window.png) Meta Prompt Extractor Mask Editor (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/mpe_mask_editor_window.png)
Occlusion Mask: Keep Foreground Objects in Front
A face swap can accidentally replace parts of a microphone, hand, glasses, or other object covering a face. Occlusion Mask creates a protection mask for those foreground pixels. It combines an occluder model, which detects what is in front of the face, with an XSeg model, which identifies face skin.
In the workflow described by the author, connect the node’s IMAGE output to ReActor’s image input and its MASK output to ReActor’s face_mask input. Connect PREVIEW to a Preview Image node to inspect the protection area before judging the face-swap result.
Key controls include:
| Control | What changing it does |
|---|---|
| Face Target | Choose the largest face for a single-person image or all faces for a group photo. |
| Face Crop Padding | Include more space around a detected face; the default is 15%. |
| Fallback to Full Image | Use the full image when no face is detected. |
| Detection Sensitivity | Increase it to catch subtler objects; decrease it if face skin is being marked. The default is 0.65. |
| Mask Expansion | Grow or shrink the protected region; the default is +6. |
| Edge Softness | Blend the mask boundary; the default is 4. |
| Mask Mode | Choose Soft for a blended boundary or Hard for a clear-cut mask. |
For a handheld microphone, the author suggests starting with sensitivity 0.65, expansion +8, softness 5, and Soft mode. Treat these as starting points: preview the mask and adjust it for the particular image.
Occlusion Mask example (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/occlusion_mask.png)
Loader for Batch Image Processing: Work with Many Images
The batch loader gathers supported images from a folder into an IMAGE batch for downstream nodes. Set image_directory to the folder you want, and enable subdirectories if images in nested folders should be included. The listed folder formats are PNG, JPG, JPEG, WebP, BMP, and GIF.
Alternatively, enable use_input_images and connect input_images to process an image batch supplied by another node. This is helpful when you want to apply the same workflow to a collection instead of loading each image individually.
PhotoLab: Give Portraits a Photographic Finish
PhotoLab combines image-wide effects with optional face-focused detail. You can make a clean portrait look grainier, softer, faded, or more textured without building a separate chain for every effect.
Its photographic controls include JPEG-style compression (quality and passes), pixelation, grain, vignette, and saturation. Color grades include None, Warm, Cool, Faded, and Sepia, with a separate strength control. Blur options include Gaussian, Box, Motion, Radial, Lens, and Soft Focus.
For portraits, PhotoLab offers controls for skin texture, pores, freckles, blemishes, acne, subsurface-scattering warmth, fine facial hair, redness, and T-zone shine. These effect strengths run from 0–100. skin_seed changes the procedural pattern, so you can explore a different texture without changing the overall look. A connected face_mask can guide face-focused work, while mask_mode offers Face Only, Inverted, and Disabled options. You can also connect a reference_image and choose a lighting-match mode.
Preset buttons provide starting points. Global presets include Film Snapshot, Darkroom B&W, Cool Editorial, Sepia Vintage, Golden Hour, Lo-Fi Degraded, and Dreamy Soft Focus. Face presets include Natural Skin, High-Detail Skin, Freckled & Rosy, Acne Breakout, Oily T-Zone, and Aged Complexion. Start with a preset, then reduce or increase individual effects to suit the portrait. The node outputs processed images and passes through a connected face_mask.
PhotoLab controls (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/PhotoLab_New.png)
Save_It: Compare, Select, and Organize Finished Images
Save_It receives images from your workflow and gives you control over which results to save and where they go. Keep AutoSave off to choose individual images manually, use Save All Images for a batch, or turn AutoSave on to save new results automatically. AutoSave deduplicates repeated frames by pixel content.
You can save PNG, JPEG, or WebP files; the Quality control applies to JPEG and WebP. A filename prefix can also specify a subfolder: Portraits/face, for example, saves numbered face files under output/Portraits/. An absolute path can target another location. Turn on Timestamp for date-time names instead of counters. Numbered files use a counter maintained separately in each save folder.
Connect an optional original_image and enable Compare Mode to review before and after images. Comparison choices include horizontal and vertical splits, an opacity-controlled overlay, and a difference view that highlights changed pixels. Save_It also provides folder favorites, an output-folder shortcut, and a history of the last 50 saved files. PNG saving preserves prompt and workflow metadata.
Save_It node (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/save_it.png) Save_It vertical-wipe comparison (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/save_it_vertical_wipe_compare.png) Save_It save history (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/save_it_save_history.png)
ialhabbal_VLLM: Work with Text, Images, and Video
The VLLM family provides Qwen and Gemma multimodal workflows, including prompt enhancement and the image-to-text and video-to-text examples shown below. In practical terms, a multimodal model can use more than plain text as input. You might use one to help turn a short concept into a fuller prompt or to produce text informed by an image.
There are standard and advanced nodes for both Hugging Face checkpoint and local GGUF workflows. The advanced nodes expose additional controls for such choices as device use, sampling, generation, or performance. Separate prompt-enhancer nodes are available for the two model-loading approaches.
VLLM prompt enhancement example (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_promptenhancer.png) VLLM image-to-text example (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_i2t.png) VLLM video-to-text example (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_v2t.png)
ComfyUI-ialhabbal Models
The VLLM nodes support Qwen and Gemma multimodal models through two loading approaches. The source documentation does not name specific model checkpoints or recommend a particular model size, so select a compatible checkpoint for the node and resources available in your workflow.
| Node | When to use it |
|---|---|
ialhabbal_VLLM | Use a Qwen or Gemma Hugging Face checkpoint for a standard multimodal workflow. |
ialhabbal_VLLM_Advanced | Use the same checkpoint approach when you need additional device, sampling, or generation controls. |
ialhabbal_VLLM_GGUF | Use a compatible local Qwen or Gemma GGUF model through llama-cpp-python. |
ialhabbal_VLLM_GGUF_Advanced | Use local GGUF inference with additional performance and configuration controls. |
ialhabbal_VLLM_PromptEnhancer | Refine or format prompts for the Hugging Face-based VLLM workflow. |
ialhabbal_VLLM_GGUF_PromptEnhancer | Refine prompts for a local GGUF-based workflow. |
Occlusion Mask also uses two task-specific models: the occluder model detects objects in front of a face, while XSeg helps distinguish face skin from those objects. They work together to produce the protection mask; they are not alternative artistic styles to choose between.
Troubleshooting ComfyUI-ialhabbal
Prompt Verify does not pause
- Check whether you connected text to
textorllm_input. Either connection loads incoming text for review and pauses the workflow. - If neither is connected and the editor already contains text, automatic submission is expected.
- If the node paused but continued before you responded, increase
timeout.
Use the Prompt Verify workflow image (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/prompt_verify_simple_workflow.png) to check the basic connections.
Meta Prompt Extractor does not return the prompt you expected
- Open the file in the browser and inspect its metadata preview.
- Confirm that the selected file contains usable prompt or workflow information. An image alone does not guarantee that prompt text is available.
- Check
use_conditioning: when it is on, connected conditioning overrides file-based prompt extraction. - For a batch, check that
batch_folderpoints to the intended folder or thatbatch_file_listcontains the exact file paths you want.
The file browser screenshot (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/mpe_Browse_Files_floating_window.png) shows where to inspect selected files.
Occlusion Mask misses an object or protects too much
- Send PREVIEW to a Preview Image node so you can see the mask itself.
- If an object has little or no coverage, raise Detection Sensitivity or Face Crop Padding.
- If the mask covers face skin, lower sensitivity or reduce Mask Expansion.
- If the result has a visible edge, try increasing expansion by 3–5 and Edge Softness by 2–3.
- If no face is found, try Fallback to Full Image. If blending produces ghosting, try Hard mask mode.
An Occlusion Mask dependency error mentions median_blur or albucore
The author identifies this as an incompatibility between installed Albumentations and Albucore versions. The stated remedy is to reinstall both in the same Python environment that runs ComfyUI, allowing the package installer to resolve a matching pair. If you do not manage that environment yourself, pass the exact error message to the person who does.
Save_It is not saving the images you intended
- Check whether AutoSave is on. Leave it off when you want to choose images manually.
- Use Save All Images if you want every image in the current batch.
- Check the filename prefix and selected folder, then use Open Output Folder to confirm the destination.
- If you need a particular file type, check Format before saving.
Frequently Asked Questions
Do I need to use all eight groups?
No. Each group exposes its own nodes, so you can add only those that help your workflow.
Can Prompt Verify give another node the text I approved?
Yes. Its STRING output contains the confirmed text. Connect CLIP if you also need its CONDITIONING output.
Can PhotoLab process more than one image?
Yes. Its images output is a processed image batch.
Can I compare a finished image with its original in Save_It?
Yes. Connect the original to original_image, then enable Compare Mode.
Will Meta Prompt Extractor recover a prompt from every image?
Not necessarily. It looks for usable metadata or workflow information in the selected file.
Learn More about ComfyUI-ialhabbal
- Browse the ComfyUI-ialhabbal project repository for the author’s source documentation and examples.
- Study the Prompt Verify workflow (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/prompt_verify_workflow.png) to see where a review step fits before generation.
- Use the Meta Prompt Extractor browser image (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/mpe_Browse_Files_floating_window.png) and Mask Editor image (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/mpe_mask_editor_window.png) as visual guides.
- Explore the VLLM examples for text-to-text (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_t2t.png), image-to-text (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_i2t.png), and video-to-text (https://raw.githubusercontent.com/ialhabbal/ComfyUI-ialhabbal/main/media/ialhabbal_vllm_v2t.png) workflows.
The project is provided under the MIT License.
