Qwen Txt2img Agent Skill

Build Qwen Image 2512 text-to-image workflows — QwenImageIntegratedKSampler, separate component loading, lightning LoRAs, and fine-tuned model variants

3k tokens

context cost

the whole folder, loaded on every use

files

instructions only

copies elsewhere

how many repositories repackaged it

481

stars on the repo

on the repository, not the skill itself

Install

one command, takes just this skill from the repository

npx skills add https://github.com/artokun/comfyui-mcp --skill qwen-txt2img

The instruction itself

24 sections, as written by the author

Qwen Image 2512 Text-to-Image Workflows

Overview

Qwen Image 2512 is the latest (December 2025) text-to-image model from the Qwen family. It uses a vision-language model (Qwen2.5-VL) as the text encoder and generates high-quality images from natural language prompts. Two workflow approaches:

QwenImageIntegratedKSampler — All-in-one node (recommended for simplicity)
Separate component loading — UNETLoader + CLIPLoader + VAELoader + standard KSampler (more flexible)

Models

Standard Components

|-----------|------|-------|-------|

Fine-tuned Variants (Installed)

| Model | Path | Focus |

|-------|------|-------|

| qwenImageEditRemix_v10 | diffusion_models/qwenImageEditRemix_v10.safetensors | General-purpose remix |

| qwenUltimateRealism_v11 | UNETLoader path | Product photography, hyper-realistic |

| copaxTimeless | UNETLoader path | Ultra-realistic portraits |

| qwnImageEdit_v16Bf16 | UNETLoader path | Abliterated (uncensored) |

Lightning LoRAs

4-Step Lightning (General Qwen / txt2img)

{
  "class_type": "LoraLoaderModelOnly",
  "inputs": {
    "model": ["<unet_node>", 0],
    "lora_name": "Qwen-Image-Lightning-4steps-V1.0.safetensors",
    "strength_model": 1.0
  }
}

Settings: steps=4, cfg=1.0, sampler=euler, scheduler=simple, denoise=1.0

8-Step Lightning (Higher Quality)

{
  "class_type": "LoraLoaderModelOnly",
  "inputs": {
    "model": ["<unet_node>", 0],
    "lora_name": "Qwen-Image-Lightning-8steps-V1.0.safetensors",
    "strength_model": 1.0
  }
}

Settings: steps=8, cfg=1.0 (or 2.5 for character detail), sampler=euler, scheduler=simple

Sampler Settings

|--------|-------|-----|---------|-----------|---------|------|-------|

ModelSamplingAuraFlow

For standard (non-lightning) presets, apply flow matching shift:

{
  "class_type": "ModelSamplingAuraFlow",
  "inputs": { "model": ["<unet_or_lora>", 0], "shift": 3.1 }
}

Shift=3.1 is the standard value for Qwen Image. Not needed with lightning LoRA (baked into the distillation).

Resolutions

Qwen operates at ~1.6 megapixels natively:

| Aspect | Resolution | Use Case |

|--------|-----------|----------|

| Square | 1328x1328 | General |

| Portrait 3:4 | 1104x1472 | Portraits |

| Portrait 2:3 | 1056x1584 | |

| Portrait 9:16 | 928x1664 | Phone format |

| Landscape 4:3 | 1472x1104 | Landscape scenes |

| Landscape 3:2 | 1584x1056 | |

| Landscape 16:9 | 1664x928 | Widescreen |

| Ultra portrait | 1536x2048 | Tall format |

| Video-ready | 832x480 | For WAN 2.2 FLF pipeline |

Approach 1: QwenImageIntegratedKSampler (All-in-One)

The QwenImageIntegratedKSampler custom node handles model patching, conditioning, sampling, and output in a single node. Simplest workflow — just 4 nodes for model loading + 1 integrated sampler + 1 save.

Node Inputs

Required:
  - model: MODEL (from UNETLoader)
  - clip: CLIP (from CLIPLoader, type=qwen_image)
  - vae: VAE
  - positive_prompt: STRING
  - negative_prompt: STRING
  - generation_mode: "文生图 text-to-image" or "图生图 image-to-image"
  - batch_size: INT (default 1)
  - width: INT (default 0, step 8)
  - height: INT (default 0, step 8)
  - seed: INT
  - steps: INT (default 4)
  - cfg: FLOAT (default 1)
  - sampler_name: euler, dpmpp_2m, etc.
  - scheduler: simple, sgm_uniform, beta, etc.
  - denoise: FLOAT (default 1)

Optional:
  - image1-5: IMAGE (reference images for i2i or multi-ref)
  - latent: LATENT
  - controlnet_data: CONTROL_NET_DATA
  - auraflow_shift: FLOAT (default 3)
  - cfg_norm_strength: FLOAT (default 1)

Outputs:
  [0] IMAGE — generated image
  [1] LATENT — output latent (optional)
  [2] IMAGE — scaled input image (for i2i)

Complete Workflow: Integrated Sampler (Lightning 4-Step)

{
  "1": { "class_type": "UNETLoader", "inputs": { "unet_name": "qwenImageEditRemix_v10.safetensors", "weight_dtype": "default" }},
  "2": { "class_type": "LoraLoaderModelOnly", "inputs": { "model": ["1", 0], "lora_name": "Qwen-Image-Lightning-4steps-V1.0.safetensors", "strength_model": 1.0 }},
  "3": { "class_type": "CLIPLoader", "inputs": { "clip_name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", "type": "qwen_image" }},
  "4": { "class_type": "VAELoader", "inputs": { "vae_name": "qwen_image_vae.safetensors" }},
  "5": { "class_type": "QwenImageIntegratedKSampler", "inputs": {
    "model": ["2", 0],
    "clip": ["3", 0],
    "vae": ["4", 0],
    "positive_prompt": "<detailed natural language prompt>",
    "negative_prompt": "",
    "generation_mode": "文生图 text-to-image",
    "batch_size": 1,
    "width": 1024,
    "height": 1344,
    "seed": 42,
    "steps": 4,
    "cfg": 1,
    "sampler_name": "euler",
    "scheduler": "simple",
    "denoise": 1,
    "auraflow_shift": 3,
    "cfg_norm_strength": 1
  }},
  "6": { "class_type": "SaveImage", "inputs": { "images": ["5", 0], "filename_prefix": "qwen_t2i" }}
}

Approach 2: Separate Component Loading (Standard Pipeline)

More flexible — allows inserting additional processing nodes between stages.

Pipeline Flow

UNETLoader → [LoraLoaderModelOnly] → [ModelSamplingAuraFlow (shift=3.1)] → MODEL
CLIPLoader (qwen_image) → CLIP
VAELoader → VAE

CLIPTextEncode (positive) → CONDITIONING
ConditioningZeroOut → negative CONDITIONING

EmptyLatentImage (1024x1344) → LATENT

KSampler → VAEDecode → SaveImage

Complete Workflow: Separate Loading (Lightning 4-Step)

{
  "1": { "class_type": "UNETLoader", "inputs": { "unet_name": "qwenImageEditRemix_v10.safetensors", "weight_dtype": "default" }},
  "2": { "class_type": "LoraLoaderModelOnly", "inputs": { "model": ["1", 0], "lora_name": "Qwen-Image-Lightning-4steps-V1.0.safetensors", "strength_model": 1.0 }},
  "3": { "class_type": "CLIPLoader", "inputs": { "clip_name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", "type": "qwen_image" }},
  "4": { "class_type": "VAELoader", "inputs": { "vae_name": "qwen_image_vae.safetensors" }},
  "5": { "class_type": "CLIPTextEncode", "inputs": { "clip": ["3", 0], "text": "<detailed natural language prompt>" }},
  "6": { "class_type": "ConditioningZeroOut", "inputs": { "conditioning": ["5", 0] }},
  "7": { "class_type": "EmptyLatentImage", "inputs": { "width": 1024, "height": 1344, "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": "qwen_t2i" }}
}

Complete Workflow: Standard Quality (50-Step)

{
  "1": { "class_type": "UNETLoader", "inputs": { "unet_name": "qwenImageEditRemix_v10.safetensors", "weight_dtype": "default" }},
  "2": { "class_type": "ModelSamplingAuraFlow", "inputs": { "model": ["1", 0], "shift": 3.1 }},
  "3": { "class_type": "CLIPLoader", "inputs": { "clip_name": "qwen_2.5_vl_7b_fp8_scaled.safetensors", "type": "qwen_image" }},
  "4": { "class_type": "VAELoader", "inputs": { "vae_name": "qwen_image_vae.safetensors" }},
  "5": { "class_type": "CLIPTextEncode", "inputs": { "clip": ["3", 0], "text": "<detailed natural language prompt>" }},
  "6": { "class_type": "ConditioningZeroOut", "inputs": { "conditioning": ["5", 0] }},
  "7": { "class_type": "EmptyLatentImage", "inputs": { "width": 1328, "height": 1328, "batch_size": 1 }},
  "8": { "class_type": "KSampler", "inputs": {
    "model": ["2", 0],
    "positive": ["5", 0],
    "negative": ["6", 0],
    "latent_image": ["7", 0],
    "seed": 42, "steps": 50, "cfg": 4, "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": "qwen_t2i_hq" }}
}

Negative Conditioning

Always use ConditioningZeroOut for Qwen txt2img:

{
  "class_type": "ConditioningZeroOut",
  "inputs": { "conditioning": ["<positive_cond>", 0] }
}

Or use an empty string in CLIPTextEncode — but ZeroOut is more explicit and reliable.

QwenImageDiffsynthControlnet

For ControlNet support with Qwen models. Patches the model with a DiffSynth control signal:

Required Inputs:
  - model: MODEL
  - model_patch: MODEL_PATCH (from DiffSynth ControlNet loader)
  - vae: VAE
  - image: IMAGE (control image)
  - strength: FLOAT (default 1.0)

Optional:
  - mask: MASK

Outputs:
  [0] MODEL (patched)

DiffSynth ControlNets support: canny, depth, inpaint only (NOT pose).

Concept/Style LoRAs (Installed)

Located in loras/Qwen/:

style/ — Figure makers, reality transform, panel painter
concept/ — Various concept LoRAs
poses/ — Pose-specific LoRAs
character/ — Character enhancement
anime/ — Anime style LoRAs
tool/ — Utility LoRAs (anything2real, gaussian splash)
equirectangular projection/ — 360 panorama LoRA

Apply with LoraLoaderModelOnly:

{
  "class_type": "LoraLoaderModelOnly",
  "inputs": {
    "model": ["<unet_or_lightning_lora>", 0],
    "lora_name": "Qwen\\concept\\hinaQwenImageAsianMixLora_v2.safetensors",
    "strength_model": 0.8
  }
}

Prompt Style

Natural language, 1–3 sentences. Be descriptive:

Good: "Professional portrait of an Asian woman in her late 20s, wearing a cream linen blazer at a Tokyo rooftop café during golden hour, holding a matcha latte, editorial fashion photography, shot on Sony A7III 85mm f/1.4"
Bad: "1girl, cafe, blazer, matcha"

Tips:

Put text to render in quotes within the prompt
"photograph" works better than "photorealistic"
Negative prompts: use NLP-style descriptions, not keyword spam (or just use ZeroOut)

VRAM Considerations

| Config | VRAM | Notes |

|--------|------|-------|

| FP8 UNET + fp8 CLIP + VAE | ~17-18GB | Fits comfortably on RTX 4090 |

| bf16 UNET (edit model) | ~10GB UNET + 7GB CLIP | Also fits well |

Always clear_vram before switching to Qwen from another model family
Lightning 4-step is extremely fast (~3-5s per image)

Tips

QwenImageIntegratedKSampler is the simplest approach for basic txt2img — one node handles everything
For LoRA stacking or ControlNet, use the separate component pipeline instead
The integrated sampler's auraflow_shift defaults to 3 (close to the recommended 3.1) — adjust only if needed
For video pipeline output (feeding into WAN FLF), set resolution to 832x480
CopaxTimeless pick: res_multistep + sgm_uniform at CFG 4.0 for ultra-realistic results
Multiple concept LoRAs can stack — reduce individual strength to 0.5-0.7 when combining

Other skills for the same job

different authors, same section of the catalogue

Canvas Design

by anthropics

vendor ×13

Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.

Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.

Improves the quality of images, especially screenshots, by enhancing resolution, sharpness, and clarity. Perfect for preparing images for presentations, documentation, or social media posts.

Downloads videos from YouTube and other platforms for offline viewing, editing, or archival. Handles various formats and quality options.

Lightweight WSI tile extraction and preprocessing. Use for basic slide processing tissue detection, tile extraction, stain normalization for H&E images. Best for simple pipelines, dataset preparation, quick tile-based analysis. For advanced spatial proteomics, multiplexed imaging, or deep learning pipelines use pathml.

Microscopy data management platform. Access images via Python, retrieve datasets, analyze pixels, manage ROIs/annotations, batch processing, for high-content screening and microscopy workflows.

Python library for working with DICOM (Digital Imaging and Communications in Medicine) files. Use this skill when reading, writing, or modifying medical imaging data in DICOM format, extracting pixel data from medical images (CT, MRI, X-ray, ultrasound), anonymizing DICOM files, working with DICOM metadata and tags, converting DICOM images to other formats, handling compressed DICOM data, or processing medical imaging datasets. Applies to tasks involving medical image analysis, PACS systems, radiology workflows, and healthcare imaging applications.

This skill should be used when working with pre-trained transformer models for natural language processing, computer vision, audio, or multimodal tasks. Use for text generation, classification, question answering, translation, summarization, image classification, object detection, speech recognition, and fine-tuning models on custom datasets.

13k tokens

How to use it

Copy the folder

Take artokun/qwen-txt2img from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.