V 1.0.7 - new readme and wf

2025-01-07 16:03:44 +01:00
parent 6e127009a3
commit 772012d53e
7 changed files with 1806 additions and 1718 deletions
@@ -4,7 +4,7 @@

 **Primere Youtube channel:** https://www.youtube.com/@PrimereComfydev/videos

-**Install 3rd party nodepack depencency:** https://github.com/city96/ComfyUI_ExtraModels
+**Install required party nodepack depencency:** https://github.com/city96/ComfyUI_ExtraModels

 <hr>

@@ -31,8 +31,10 @@ After nodepack update of **12/3/2024 - v1.0.2** ComfyUI front-end must be update
 - Prompt encoder with selectable custom clip model, long-clip mode with custom models, advanced encoding, injectable internal styles, last-layer options
 - Sampler with `variation extender` and `Align Your Step` features
 - A1111 style network injection supported by text prompt (Lora, Lycorys, Hypernetwork, Embedding)
- Automatized and manual image saver. Manual image saver with optional **preview saver** for checkpoint selector and saved .csv prompts
+- Automatized and manual image saver. Manual image saver with optional **preview saver** for checkpoint (Lora, Lycoris, Embedding) selectors and saved .csv prompts
 - Upscaler (selectable Ultimate SD and hiresFix)
+- Dynamic prompt support
+- Auto clean incompatible network tags from prompt by model arhitechture

 <hr>

@@ -49,7 +51,7 @@ The main difference between **minimal** and **basic** workflows, that **basic**
 #### Same as Minimal workflow plus:

 - **Half-automatic model concept selector:**
-  - **Supported concepts:** SD1, SD2, SDXL, SD3, StableCascade, Turbo, Flux, KwaiKolors, Hunyuan, Playground, Pony, LCM, Lightning, Hyper, PixartSigma, Sana (both 1024 and 512)
+  - **Supported concepts:** SD1, SD2, SDXL, SD3, StableCascade, Turbo, Flux, KwaiKolors, Hunyuan DiT (image only), Playground, Pony, LCM, Lightning, Hyper, PixartSigma, Sana (both 1024 and 512)
  - Custom (and different) sampler settings for all concepts. The main idea is set sampler nodes only one time (`sampler`, `scheduler`, `step`, `cfg`) then just select model only what will use right sampler, vae, clip settings by `Model concept selector`. 
  - Auto detection of selected model type (if data already stored on external .json file, see longer [manual](Workflow/Manual/nodes/basic_workflow.md))
  - Auto **download** and apply Hyper, Lightning, and Turbo speed loras at first usage from here: https://huggingface.co/ByteDance/Hyper-SD/tree/main **check your SSD space before!**
@@ -72,8 +74,8 @@ The main difference between **basic production** and **basic** workflows, that *

 - Added 4 test and 4 development prompt inputs, easy to switch
 - Local LLM models can help refine/repair prompts. Refined prompts can be added to original, replace original, or keep original as L prompt but send refined to T5-XXL clip is avalable in clip encoder node 
- Customizable refiner blocks for face, eye, mouth, and hand refining. These nodes automatically downloads required segmentator models, check space before first usage
- Refiner blocks using [DeepFace analyzer](https://github.com/serengil/deepface) if needed
+- Customizable refiner blocks for face, eye, mouth, and hand refining. Auto segmentation model downloads deleted, manual model download required
+- Refiner blocks using [DeepFace analyzer](https://github.com/serengil/deepface) if needed, detect age, race, gender and mood

 <hr>

@@ -151,7 +151,7 @@ Remember: The effectiveness of prompt enhancement depends on both model selectio

 ## Components Overview:

-This powerful image refinement system consists of three specialized nodes working together to provide precise control over image enhancement:
+This powerful image refinement system consists of four specialized nodes working together to provide precise control over image enhancement:

 1. Primere Refiner Prompt
 2. Primere Image Segments
@@ -167,6 +167,7 @@ Test workflow: `[your_comfy_dir]\custom_nodes\ComfyUI_Primere_Nodes\Workflow\civ
 - Model and VAE selection for refiners only
 - Network adapter support (LoRA, LyCORIS, Embeddings, Hypernetworks)
 - Concept-specific processing controls (you can on off refining process for Flux, Cascade, or any other supported model concepts)
+- DeepFace analyzer folow age, race, gender and mood if required models avalilable

 ### Positive/Negative Prompting for refiners:
 ```plaintext
@@ -208,11 +209,12 @@ You can mix the original positive and negative prompts to refining process with
 **This Python module change another modules by version. Make backup of your current Python libs before install DeepFace. If something wrong after installed DeepFace, just revert back the updated Python libs to previous state from backup.**

 #### DeepFace installation:
- Backup your current Pythin libraries
- Use `pip install deepface` on terminal, but dont't forget to activate virtual environment
- Start comfy, and if failed bcause library versions changed, just copy back the original version from backup
+- Backup your current Python libraries
+- Use `pip install deepface` on terminal, but dont't forget to activate virtual environment before
+- Start comfy, and if failed because Python library versions changed, just copy back the original version from backup
 - Download 4 required weights: `'age_model_weights.h5', 'facial_expression_model_weights.h5', 'gender_model_weights.h5', 'race_model_single_batch.h5'` from here: https://github.com/serengil/deepface_models/releases/ and save them to this folder: `[comfypath]\models\deepface\.deepface\weights\`
 - Use four On/Off switches of analyzer to compare results
+- Check difference between on/off states using same prompt and seed

 #### Examples:
 Face detailer without analyzer:
@@ -327,6 +329,8 @@ Remember: Fine-tune these settings based on your specific use case and desired o
 #### Hands, but aesthetic scorer automatically ignored the second:
 <img src="refiner_block_hands2.jpg" width="500px">

+#### Auto model downloader deleted, you must download segmentation models manually and save them to right path. [Here are examples what to download and where to save](segment_model_download.md)
+
 <hr>

 # <ins>DiT Prompt Purifier Node:</ins>
@@ -344,7 +348,7 @@ The DiT Prompt Purifier node automatically cleans and optimizes prompts for mode
 - Selective cleaning based on model Architecture
 - Incompatible weight and structure removal

-### Cleaning Operations
+### Cleaning Operations:

 #### Removes:
 - Prompt weights (e.g., `(element:1.2)`)
@@ -259,12 +259,12 @@ This sophisticated prompt encoder node offers extensive control over prompt proc
 - `clip_mode`: Toggle between standard CLIP and Long-CLIP processing
  - `clip_model/longclip_model`: Model selection based on clip_mode switch
 - `last_layer`: Fine-tune CLIP encoding by selecting specific negative layers for feature extraction
- `negative_strength`: Adjusts the intensity of negative prompt influence
+- `negative_strength`: Adjusts the intensity of negative prompt influence (if the selected model support)

 ### Style System:
 - `use_int_style`: Enables internal style system loaded from external .toml configurations from the path: `Toml/default_neg.toml` and `Toml/default_pos.toml`
-  - `int_style_pos/neg`: Select predefined styles by name
-  - `int_style_pos/neg_strength`: Control strength of applied styles
+  - `int_style_pos/int_style_neg`: Select predefined styles by name
+  - `int_style_pos_strengt/int_style_neg_strength`: Control strength of applied styles

 ### Advanced Encoding Options:
 - `adv_encode`: Enables alternative (advanced) CLIP encoding methodology
@@ -279,11 +279,16 @@ This sophisticated prompt encoder node offers extensive control over prompt proc
  - `T5-XXL`: Uses enhanced prompt input for T5-XXL encoding if concept support T5 clipping
 - `enhanced_prompt_strength`: Controls enhanced prompt influence if added to original positive prompt

-### Additional Controls:
- `style_position`: Placement of additional style prompts (Top/Bottom)
-  - `opt_pos/neg_strength`: Fine-tune optional prompt strengths
- `copy_prompt_to_l`: Enables SDXL first-pass prompt copying
-  - `sdxl_l_strength`: Controls SDXL first-pass prompt intensity
+### Additional Style Controls:
+- `style_handling`: Separate or merge styles with original prompt
+#### If style and prompt merged:
+- `style_position`: Placement of additional style (Top/Bottom)
+#### If style and prompt separated: 
+- `style_swap`: Style send to default clip, prompt send to T5/L encoder or Style send to T5/L, prompt send to default clip
+
+### Optional prompt handling:
+- `opt_pos_strength/opt_neg_strength`: Fine-tune optional prompt strengths
+- `l_strength`: Controls SDXL first-pass prompt intensity id L prompt connected to node

 <hr>

@@ -0,0 +1,75 @@
+# <ins>Model sources and destionations for Primere prompt enhancer:</ins>
+
+### <ins>BBOX Sources:</ins>
+
+```
+    https://huggingface.co/Bingsu/adetailer/resolve/main/face_yolov8m.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/face_yolov8n.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/face_yolov8n_v2.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/face_yolov8s.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/hand_yolov8n.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/hand_yolov8s.pt?download=true
+    https://huggingface.co/ultralyticsplus/yolov8s/resolve/main/yolov8s.pt?download=true
+```    
+save them to:
+```    
+    models\ultralytics\bbox\
+```
+
+<hr>
+
+### <ins>SEGM Sources:</ins>
+
+```
+    https://huggingface.co/Bingsu/adetailer/resolve/main/deepfashion2_yolov8s-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/face_yolov8m-seg_60.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/face_yolov8n-seg2_60.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/facial_features_yolo8x-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/flowers_seg_yolov8model.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/hair_yolov8n-seg_60.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/person_yolov8m-seg.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/person_yolov8n-seg.pt?download=true
+    https://huggingface.co/Bingsu/adetailer/resolve/main/person_yolov8s-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/skin_yolov8m-seg_400.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/skin_yolov8n-seg_400.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/skin_yolov8n-seg_800.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8l-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8m-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8n-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8s-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8x-seg.pt?download=true
+    https://huggingface.co/jags/yolov8_model_segmentation-set/resolve/main/yolov8_butterfly_custom.pt?download=true
+```
+save them to:
+```    
+    models\ultralytics\segm\
+```
+
+<hr>
+
+### <ins>DINO Sources:</ins>
+
+```
+    https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swinb_cogcoor.pth?download=true
+    https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/GroundingDINO_SwinB.cfg.py?download=true
+    https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swint_ogc.pth?download=true
+    https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/GroundingDINO_SwinT_OGC.cfg.py?download=true
+```
+save them to:
+```    
+    models\grounding-dino\
+```
+
+<hr>
+
+### <ins>SAMS Sources:</ins>
+
+```
+    https://huggingface.co/ybelkada/segment-anything/resolve/main/checkpoints/sam_vit_b_01ec64.pth?download=true
+    https://huggingface.co/ybelkada/segment-anything/resolve/main/checkpoints/sam_vit_h_4b8939.pth?download=true
+    https://huggingface.co/ybelkada/segment-anything/resolve/main/checkpoints/sam_vit_l_0b3195.pth?download=true
+```
+save them to:
+```    
+    models\sams\
+```