GeekatplayStudio-ComfyUI-cluster/docs/WORKFLOWS.md

Workflows

1) Text -> Checkpoint + LoRA

File: workflows/ollama_text_checkpoint_lora.json

Flow:

• OllamaPromptPlanner
• DynamicCheckpointLoader
• DynamicLoraStack
• CLIPTextEncode (positive/negative)
• KSampler
• VAEDecode
• SaveImage

Use when:

• prompt-only generation
• standard SDXL/SD1.5 checkpoints
• automatic LoRA selection from planner output

2) Text -> Flux Split + LoRA

File: workflows/ollama_text_flux_split_lora.json

Flow differences:

• planner uses task_hint = flux
• loader uses model_type = flux
• loader receives planner vae_name and clip_name as overrides

Use when:

• flux diffusion model is separate from VAE and text encoders
• prompt-only generation with optional LoRA stack

3) Image + Prompt -> Checkpoint + LoRA

File: workflows/ollama_image_checkpoint_lora.json

Flow:

• LoadImage
• OllamaVisionStylePlanner analyzes image + prompt
• standard checkpoint and LoRA path
• VAEEncode for img2img latent
• KSampler with planner denoise

Use when:

• image-guided stylization or enhancement with non-flux models

4) Image + Prompt -> Flux Split + LoRA

File: workflows/ollama_image_flux_split_lora.json

Flow differences:

• vision planner with task_hint = flux
• flux split overrides for clip/vae
• img2img latent path through VAEEncode

Use when:

• image-guided generation with flux diffusion models and separate clip/vae files

Debug/Status

Each workflow includes LiveStatus nodes to show:

• planner JSON
• selected checkpoint or diffusion model
• loader debug output
• LoRA stack debug output