Lesson 15: Full ComfyUI Workflows for Consistency & Batch Production

Lesson 15 concludes the basics of image generation mastery by teaching how to build, save, load, and optimize complete reusable ComfyUI workflows. These workflows enable consistent character creation, batch generation of variations, automated testing of prompts/settings, and scalable production — turning single images into professional series or portfolios.

Core Benefits of Saved & Modular Workflows

• Reproducibility: Load a JSON file → same base quality every time
• Consistency: Fix character face/body across dozens of poses/scenes
• Batch efficiency: Generate 20–100 variations automatically
• Testing & iteration: Use XY Plot to compare prompts, CFG, steps, LoRAs visually
• Modularity: Swap models/LoRAs/ControlNets without rebuilding everything

Building a Pro Reusable Workflow Template

Use your Lesson 10 basic txt2img as foundation and expand it into a full professional template.

1. Core Chain:
   • Load Checkpoint (HiDream-I1 uncensored)
   • Load LoRA nodes (stack 2–3: skin, anatomy, wetness)
   • CLIP Text Encode (positive & negative)
   • Empty Latent Image (1024×1536 portrait)
   • KSampler (60 steps, CFG 5.2, DPM++ 2M Karras)
   • VAE Decode → Save Image
2. Add ControlNet Branch (from Lesson 12):
   • Load Image (reference pose)
   • ControlNet Preprocessor (OpenPose + Depth)
   • Apply ControlNet (strength 1.0 OpenPose, 0.85 Depth)
   • Connect to KSampler control_net input
3. Optional IPAdapter/FaceID (for character consistency):
   • Load reference face image
   • IPAdapter Apply node → connect to model/CLIP
4. Batch & Variation Controls:
   • Set batch size in Empty Latent or use Batch Prompt node
   • Add Random Seed node for variation

XY Plot for Prompt/Settings Testing

1. Install "Efficiency Nodes" or "ComfyUI-Impact-Pack" if not already (via Manager).
2. Add XY Plot node (from Efficiency or Impact-Pack).
3. Configure:
   • X-axis: CFG scale (4.5 to 6.0, step 0.5)
   • Y-axis: Steps (40 to 80, step 10)
   • Or X: LoRA strength (0.5 to 1.0), Y: different prompts
4. Connect XY Plot output to KSampler → generate grid.
5. Review grid: Identify sweet spot for sharpness, naturalness, artifact reduction.

Batch Generation & Queue Management

• Set batch size in Empty Latent: 8–16 for variations
• Use Seed Random node + increment for unique outputs
• Queue multiple prompts: Add Prompt List node or manual queue
• Monitor VRAM: Batch too high → reduce resolution or use quantized model

Saving, Organizing & Loading Workflows

1. Right-click canvas → Save → name e.g., "Pro_Photoreal_HiDream_Control_LoRA.json"
2. Create folder structure: ComfyUI/workflows/
   • basics/
   • controlnet/
   • loras/
   • upscale_inpaint/
   • character_consistency/
3. Load: Drag JSON into canvas or File → Load
4. Version control: Add date/model suffix to filenames

Assignment

1. Build a full pro template workflow incorporating:
   • HiDream-I1 uncensored
   • 2–3 LoRAs (skin, anatomy, wetness)
   • OpenPose + Depth ControlNet
   • Batch size 8
2. Save as "Pro_Template_v1.json"
3. Run 2 batches with Lesson 10 refined prompt → generate 16 images total
4. Use XY Plot to test 2 variables (e.g., CFG 4.8–5.8 and LoRA strength 0.7–0.9)
5. Save the grid output and select top 4–6 images
6. Document: Which combination gave the best balance of detail, realism, and consistency?

This workflow template is now your production engine. Future lessons build specialized variants for character consistency, complex scenes, video motion, and portfolio-scale output.

End of Lesson 15

Chapter 2: Image Generation Mastery – Basics is now complete. The next chapter advances to character consistency, multi-subject scenes, advanced animation, optimization, and capstone projects.