Builds Flux txt2img workflows using Flux.1 Dev SRPO, Flux 2 Klein 9B, Turbo LoRAs, FluxGuidance, DualCLIPLoader, and CLIPTextEncodeFlux nodes.
From comfynpx claudepluginhub artokun/comfyui-mcp --plugin comfyThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Enables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Flux is a guidance-distilled diffusion model family from Black Forest Labs. It uses a separate FluxGuidance node instead of KSampler CFG (which must always be 1.0). Three variants are available locally:
flux2-vae.safetensors. Fast 4-step generation.| Component | Node | Model | Notes |
|---|---|---|---|
| UNET | UNETLoader | flux.1-dev-SRPO-BFL-bf16.safetensors | 22.7GB, BF16 only — FP8 produces broken results |
| CLIP | DualCLIPLoader (type=flux) | clip_name1: t5xxl_fp8_e4m3fn.safetensors, clip_name2: clip_l.safetensors | T5XXL (4.7GB) + CLIP-L (235MB) |
| VAE | VAELoader | ae.safetensors | Standard Flux/Z-Image VAE (320MB) |
| Component | Node | Model | Notes |
|---|---|---|---|
| UNET | UNETLoader | bigLove_klein1.safetensors | 17.3GB, Klein 9B variant |
| CLIP | CLIPLoader (type=flux) | qwen_3_8b_fp8mixed.safetensors | Qwen3-8B in text_encoders/ (8.3GB) |
| VAE | VAELoader | flux2-vae.safetensors | Flux 2 specific VAE (321MB) |
Klein 9B vs Flux.1 Dev: Klein uses Qwen3-8B text encoder (not T5XXL + CLIP-L). It has a different VAE (flux2-vae.safetensors). 9B distilled runs in 4 steps; 9B base needs ~50 steps at CFG 5.0. Fits in ~20GB VRAM with FP8.
| Component | Node | Model | Notes |
|---|---|---|---|
| LoRA | LoraLoaderModelOnly | flux2-turbo-lora.safetensors | 2.6GB, strength 1.0 |
| Alt LoRA | LoraLoaderModelOnly | Flux2TurboComfyv2.safetensors | Community variant, same size |
Provides separate prompt fields for each text encoder:
{
"class_type": "CLIPTextEncodeFlux",
"inputs": {
"clip": ["<dual_clip>", 0],
"clip_l": "short prompt for CLIP-L",
"t5xxl": "detailed description for T5XXL",
"guidance": 3.5
}
}
clip_l captures key semantic features. t5xxl expands and refines descriptions. For simple use, put the same prompt in both fields. Guidance is built into this node — no separate FluxGuidance needed.
If using standard CLIPTextEncode instead of CLIPTextEncodeFlux, apply guidance separately:
{
"class_type": "FluxGuidance",
"inputs": {
"conditioning": ["<clip_text_encode>", 0],
"guidance": 3.5
}
}
| Scenario | Guidance | Notes |
|---|---|---|
| Short prompts | 3.5–4.0 | Tighter prompt adherence |
| Long/complex prompts | 1.0–1.5 | More creative freedom |
| Realism | 2.5 | Less glossy skin, richer detail |
| Standard | 3.5 | Default for most use cases |
Flux does NOT support traditional negative prompts (guidance-distilled, CFG=1.0). Use ConditioningZeroOut:
{
"class_type": "ConditioningZeroOut",
"inputs": { "conditioning": ["<positive_cond>", 0] }
}
Or simply use an empty CLIPTextEncode for the negative input.
| Parameter | Standard | Notes |
|---|---|---|
| steps | 20 | Range: 20–28 |
| cfg | 1.0 | Always 1.0 — guidance is via FluxGuidance |
| sampler_name | ipndm | Author-recommended for SRPO |
| scheduler | beta | Author-recommended for SRPO |
| guidance | 3.5 | Via CLIPTextEncodeFlux or FluxGuidance |
| denoise | 1.0 |
SRPO note: The ipndm/beta combo is specifically recommended by the SRPO author. Standard Flux settings (euler/simple) also work but ipndm/beta gives better results with this fine-tune.
| Parameter | Value | Notes |
|---|---|---|
| steps | 4 | Distilled model, 4 steps is optimal |
| cfg | 1.0 | Always 1.0 |
| sampler_name | euler | |
| scheduler | simple | |
| denoise | 1.0 |
| Parameter | Value | Notes |
|---|---|---|
| steps | 50 | Full quality |
| cfg | 5.0 | Higher CFG for base model |
| sampler_name | euler | |
| scheduler | simple |
| Parameter | Value | Notes |
|---|---|---|
| steps | 4 | Turbo-distilled |
| cfg | 1.0 | |
| sampler_name | euler | |
| scheduler | simple | |
| lora_strength | 1.0 |
| Aspect | Resolution | Megapixels |
|---|---|---|
| Square | 1024x1024 | 1.0MP |
| Portrait 3:4 | 896x1152 | 1.0MP |
| Landscape 4:3 | 1152x896 | 1.0MP |
| Landscape 16:9 | 1344x768 | 1.0MP |
| Portrait 9:16 | 768x1344 | 1.0MP |
Flux operates at ~1 megapixel natively. Dimensions should be multiples of 8.
Natural language descriptions. No quality tags needed (unlike SDXL/Illustrious). Detailed, descriptive prompts work best.
Good: "A young woman with auburn hair sits at a sunlit cafe in Paris, wearing a cream linen blazer, soft bokeh background, shot on Sony A7III 85mm f/1.4"
Bad: "masterpiece, best quality, 1girl, cafe, paris"
{
"1": { "class_type": "UNETLoader", "inputs": { "unet_name": "flux.1-dev-SRPO-BFL-bf16.safetensors", "weight_dtype": "default" }},
"2": { "class_type": "DualCLIPLoader", "inputs": { "clip_name1": "t5xxl_fp8_e4m3fn.safetensors", "clip_name2": "clip_l.safetensors", "type": "flux" }},
"3": { "class_type": "VAELoader", "inputs": { "vae_name": "ae.safetensors" }},
"4": { "class_type": "CLIPTextEncodeFlux", "inputs": {
"clip": ["2", 0],
"clip_l": "<short prompt>",
"t5xxl": "<detailed prompt>",
"guidance": 3.5
}},
"5": { "class_type": "ConditioningZeroOut", "inputs": { "conditioning": ["4", 0] }},
"6": { "class_type": "EmptyLatentImage", "inputs": { "width": 896, "height": 1152, "batch_size": 1 }},
"7": { "class_type": "KSampler", "inputs": {
"model": ["1", 0],
"positive": ["4", 0],
"negative": ["5", 0],
"latent_image": ["6", 0],
"seed": 42, "steps": 20, "cfg": 1, "sampler_name": "ipndm", "scheduler": "beta", "denoise": 1
}},
"8": { "class_type": "VAEDecode", "inputs": { "samples": ["7", 0], "vae": ["3", 0] }},
"9": { "class_type": "SaveImage", "inputs": { "images": ["8", 0], "filename_prefix": "flux_srpo" }}
}
{
"1": { "class_type": "UNETLoader", "inputs": { "unet_name": "bigLove_klein1.safetensors", "weight_dtype": "default" }},
"2": { "class_type": "CLIPLoader", "inputs": { "clip_name": "qwen_3_8b_fp8mixed.safetensors", "type": "flux" }},
"3": { "class_type": "VAELoader", "inputs": { "vae_name": "flux2-vae.safetensors" }},
"4": { "class_type": "CLIPTextEncode", "inputs": { "clip": ["2", 0], "text": "<prompt>" }},
"5": { "class_type": "ConditioningZeroOut", "inputs": { "conditioning": ["4", 0] }},
"6": { "class_type": "EmptyLatentImage", "inputs": { "width": 1024, "height": 1024, "batch_size": 1 }},
"7": { "class_type": "KSampler", "inputs": {
"model": ["1", 0],
"positive": ["4", 0],
"negative": ["5", 0],
"latent_image": ["6", 0],
"seed": 42, "steps": 4, "cfg": 1, "sampler_name": "euler", "scheduler": "simple", "denoise": 1
}},
"8": { "class_type": "VAEDecode", "inputs": { "samples": ["7", 0], "vae": ["3", 0] }},
"9": { "class_type": "SaveImage", "inputs": { "images": ["8", 0], "filename_prefix": "flux_klein" }}
}
Klein note: Uses single CLIPLoader (not DualCLIPLoader) with type: "flux" and the Qwen3-8B text encoder from text_encoders/. The CLIP loader path resolves from models/text_encoders/.
{
"1": { "class_type": "UNETLoader", "inputs": { "unet_name": "flux.1-dev-SRPO-BFL-bf16.safetensors", "weight_dtype": "default" }},
"2": { "class_type": "LoraLoaderModelOnly", "inputs": { "model": ["1", 0], "lora_name": "flux2-turbo-lora.safetensors", "strength_model": 1.0 }},
"3": { "class_type": "DualCLIPLoader", "inputs": { "clip_name1": "t5xxl_fp8_e4m3fn.safetensors", "clip_name2": "clip_l.safetensors", "type": "flux" }},
"4": { "class_type": "VAELoader", "inputs": { "vae_name": "ae.safetensors" }},
"5": { "class_type": "CLIPTextEncodeFlux", "inputs": {
"clip": ["3", 0],
"clip_l": "<short prompt>",
"t5xxl": "<detailed prompt>",
"guidance": 3.5
}},
"6": { "class_type": "ConditioningZeroOut", "inputs": { "conditioning": ["5", 0] }},
"7": { "class_type": "EmptyLatentImage", "inputs": { "width": 1024, "height": 1024, "batch_size": 1 }},
"8": { "class_type": "KSampler", "inputs": {
"model": ["2", 0],
"positive": ["5", 0],
"negative": ["6", 0],
"latent_image": ["7", 0],
"seed": 42, "steps": 4, "cfg": 1, "sampler_name": "euler", "scheduler": "simple", "denoise": 1
}},
"9": { "class_type": "VAEDecode", "inputs": { "samples": ["8", 0], "vae": ["4", 0] }},
"10": { "class_type": "SaveImage", "inputs": { "images": ["9", 0], "filename_prefix": "flux_turbo" }}
}
Apply Flux LoRAs with LoraLoaderModelOnly between UNET and KSampler:
{
"class_type": "LoraLoaderModelOnly",
"inputs": {
"model": ["<unet_or_previous_lora>", 0],
"lora_name": "<lora_file>.safetensors",
"strength_model": 1.0
}
}
Klein 9B LoRAs go in loras/Flux.2 Klein 9B/ subfolder:
klein_slider_detail.safetensors — Detail slider LoRA| Model | VRAM | Notes |
|---|---|---|
| SRPO BF16 + DualCLIP | ~24GB | Fills RTX 4090 exactly. Must use BF16 — FP8 is broken for SRPO |
| Klein 9B FP8 + Qwen3-8B | ~20GB | Fits comfortably on 4090 |
| SRPO + Turbo LoRA | ~24GB | Same as SRPO base |
clear_vram before switching to Flux from another model familyCLIPTextEncodeFlux or FluxGuidance