Image Generation - Comput3 Network

Generate high-quality images using ComfyUI workflows on Comput3 Network’s GPU infrastructure. Simply launch a GPU instance and start creating with powerful node-based workflows.

Available Image Generation Models

Choose from a wide variety of specialized image generation models:

Qwen-Image Text to Image

Generate images with exceptional multilingual text rendering and editing capabilities

Model: Qwen-Image’s 20B MMDiT model
Specialty: Multilingual text rendering, advanced editing
Best for: Text-heavy images, multilingual content, detailed editing

OmniGen2 Text to Image

Generate high-quality images from text prompts using unified multimodal model

Model: OmniGen2’s unified 7B multimodal model
Architecture: Dual-path architecture
Best for: High-quality text-to-image generation, versatile applications

OmniGen2 Image Edit

Edit images with natural language instructions

Model: OmniGen2’s advanced image editing capabilities
Features: Text rendering support, natural language editing
Best for: Image modification, style changes, content editing

Cosmos Predict2 2B T2I

Generate physically accurate, high-fidelity images

Model: Cosmos-Predict2 2B T2I
Specialty: Physically accurate, detail-rich generation
Best for: Realistic images, scientific visualization, detailed artwork

Chroma Text to Image

Modified Flux architecture for enhanced image generationModel: Chroma (modified from Flux)
Architecture: Enhanced Flux-based architecture
Best for: High-quality generation, architectural improvements

HiDream Series

Multiple HiDream models for different use cases

HiDream I1 Dev: Development and testing
HiDream I1 Rapide: Fast image generation
HiDream I1 Complet: Full-featured generation
HiDream E1.1 Image Edit: Advanced image editing (better quality than E1)
HiDream E1 Image Edit: Image editing capabilities

Stable Diffusion 3.5 Series

Latest Stable Diffusion with advanced features

SD3.5 Simple: Standard text-to-image generation
SD3.5 Grand Canny ControlNet: Edge detection guided generation
SD3.5 Grande Profondeur: Depth-aware image generation
SD3.5 Grand Flou: Blur-based reference image generation

Stable Diffusion XL Series

High-quality SDXL models with various capabilities

SDXL Simple: High-quality standard generation
SDXL Refiner Prompt: Enhanced results with refiners
Révisions de Texte SDXL: Reference image concept transfer
Révision Zéro Positive SDXL: Text prompts with reference images
SDXL Turbo: Single-step image generation

Lotus Depth

Zero-shot monocular depth estimation

Model: Lotus Depth in ComfyUI
Specialty: Efficient depth estimation with high detail retention
Best for: Depth-aware applications, 3D processing

Quick Start

Launch GPU Instance

Launch a GPU instance with the ComfyUI template from your Comput3 dashboard.

ComfyUI Template

Pre-configured instance with ComfyUI, popular models, and workflows ready to use.

Recommended GPU

RTX 4090 48GB, L40S, or A100 for optimal performance with image generation workflows.

Access ComfyUI

Connect to your GPU instance and open the ComfyUI web interface.

# ComfyUI runs on port 8188 by default
http://<your-instance-ip>:8188

The ComfyUI template automatically starts the service and makes it accessible via web browser.

Choose a Workflow

Select from pre-installed workflows or create your own:

Text-to-Image
Image-to-Image
Advanced Workflows

Generate images from text descriptions

Stable Diffusion XL workflows
ControlNet integration
LoRA model support
Batch generation capabilities

Generate and Download

Run your workflow in ComfyUI and download the generated images.

Images are saved locally on your GPU instance and can be downloaded via the ComfyUI interface or SSH.

Advanced Prompt Techniques

Prompt Structure Best Practices

Subject Description

Be specific about the main subject:❌ Vague: “A person” ✅ Specific: “A middle-aged woman with curly red hair wearing a blue dress”❌ Generic: “A building”
✅ Detailed: “A modern glass skyscraper with geometric patterns”

Action and Pose

Describe what the subject is doing:

“running through a field”
“sitting peacefully by a window”
“dancing in the rain”
“looking directly at the camera”
“reaching toward the sky”

Environment and Setting

Set the scene and atmosphere:

“in a mystical forest with glowing mushrooms”
“on a busy city street at night”
“in a cozy library with warm lighting”
“against a dramatic storm sky”
“in a minimalist white studio”

Style and Aesthetic

Specify the artistic style:Photography Styles:

“professional portrait photography”
“street photography, candid moment”
“macro photography, extreme close-up”
“aerial photography, bird’s eye view”

Art Styles:

“oil painting in impressionist style”
“digital art, concept art style”
“watercolor illustration, soft edges”
“pencil sketch, detailed line art”

Technical Parameters

Add technical quality descriptors:

“8k ultra high resolution”
“cinematic lighting, dramatic shadows”
“shallow depth of field, bokeh background”
“HDR, vibrant colors, high contrast”
“soft natural lighting, golden hour”

Prompt Weighting and Control

Attention Weighting
Negative Prompts
Style Mixing

Control emphasis with parentheses and weights:

(beautiful sunset:1.3) over (mountain lake:1.1), 
(dramatic clouds:0.9), peaceful atmosphere

Weight Guidelines:

(keyword:1.3) - Increase emphasis by 30%
(keyword:0.8) - Decrease emphasis by 20%
((keyword)) - Strong emphasis (equivalent to 1.21)
[keyword] - Slight de-emphasis (equivalent to 0.91)

Exclude unwanted elements:

Prompt: "Portrait of a woman, professional photography"

Negative: "blurry, low quality, distorted face, bad anatomy, 
extra limbs, watermark, signature, text"

Common Negative Prompts:

Quality issues: blurry, low quality, pixelated, grainy
Anatomy problems: bad anatomy, extra limbs, deformed hands
Unwanted elements: watermark, signature, text, logo
Style issues: cartoon, anime (if you want photorealistic)

Combine multiple artistic influences:

"Portrait in the style of (Renaissance painting:0.7) mixed with 
(modern photography:0.5), chiaroscuro lighting, classical composition"

Effective Combinations:

Photography + Painting: photorealistic portrait, oil painting texture
Multiple artists: style of Van Gogh and Monet, impressionist fusion
Era mixing: Art Deco design with modern minimalism

Model-Specific Guidelines

Stable Diffusion XL (SDXL)

Strengths

Exceptional detail and resolution
Great with complex compositions
Excellent text rendering in images
Superior photorealism capabilities

Optimal Settings

Steps: 25-35 (30 recommended)
Guidance Scale: 6-9 (7.5 recommended)
Resolution: 1024x1024 or 1152x896
Sampler: DPM++ 2M Karras

Best Prompting Practices:

# Good SDXL prompt structure
"[Detailed subject description], [specific artistic style], 
[lighting conditions], [composition notes], [quality modifiers]"

# Example
"Close-up portrait of an elderly craftsman with weathered hands, 
Renaissance painting style, dramatic chiaroscuro lighting, 
three-quarter view composition, oil painting texture, masterpiece quality"

Stable Diffusion 2.1

Strengths

Fast generation times
Good for iteration and experimentation
Reliable results with simple prompts
Cost-effective for batch generation

Optimal Settings

Steps: 20-30 (25 recommended)
Guidance Scale: 7-12 (9 recommended)
Resolution: 512x512 or 768x512
Sampler: Euler a or DPM++ 2M

Best Prompting Practices:

# SD 2.1 responds well to clear, direct prompts
"[Subject], [style], [lighting], [quality]"

# Example  
"Mountain landscape, digital art style, golden hour lighting, highly detailed"

DALL-E Style Model

Strengths

Excellent instruction following
Great with complex scene descriptions
Superior text integration
Photorealistic human faces

Optimal Settings

Steps: 30-40 (35 recommended)
Guidance Scale: 8-12 (10 recommended)
Resolution: 1024x1024, 1024x1792, 1792x1024
Sampler: DDIM or DPM++ SDE

Best Prompting Practices:

# DALL-E style model excels with natural language descriptions
"A photograph of [detailed scene description] captured with 
[camera/lens details], [lighting conditions], [mood/atmosphere]"

# Example
"A photograph of a cozy coffee shop interior during a rainy afternoon, 
captured with a 35mm lens, warm ambient lighting filtering through 
large windows, creating a peaceful and inviting atmosphere"

Generation Parameters

Resolution and Aspect Ratios

Square Formats
Portrait Formats
Landscape Formats

1:1 Aspect Ratio

Resolution	Use Case	Cost
512x512	Social media avatars, icons	1x
768x768	Instagram posts, thumbnails	1.5x
1024x1024	High-quality social media, prints	2x
1536x1536	Large prints, detailed artwork	3x

Resolution	Aspect Ratio	Use Case
512x768	2:3	Portrait photography
768x1024	3:4	Instagram stories, mobile
1024x1536	2:3	Professional portraits
1024x1792	9:16	Vertical social media

Resolution	Aspect Ratio	Use Case
768x512	3:2	Traditional photography
1024x768	4:3	Computer wallpapers
1344x768	16:9	Widescreen, YouTube thumbnails
1792x1024	16:9	Banner images, headers

Quality vs Speed Settings

Draft Quality (Fast)

For rapid iteration and concept exploration

Steps: 15-20
Guidance Scale: 6-8
Generation Time: 1-2 seconds
Cost: Standard pricing

Best for: Brainstorming, quick concepts, batch generation

Standard Quality (Balanced)

For most production use cases

Steps: 25-35
Guidance Scale: 7-9
Generation Time: 2-4 seconds
Cost: Standard pricing

Best for: Social media, web use, general content creation

High Quality (Detailed)

For professional and print applications

Steps: 40-50
Guidance Scale: 8-12
Generation Time: 4-8 seconds
Cost: 1.5x standard pricing

Best for: Print materials, professional work, detailed artwork

Ultra Quality (Maximum)

For the highest quality results

Steps: 60-80
Guidance Scale: 10-15
Generation Time: 8-15 seconds
Cost: 2x standard pricing

Best for: Gallery prints, commercial use, artistic masterpieces

Batch Generation and Variations

Generating Multiple Images

import requests

def generate_batch_images(prompt, count=4):
    response = requests.post(
        "https://api.comput3.ai/v1/generate/image",
        headers={"Authorization": "Bearer YOUR_API_KEY"},
        json={
            "model": "stable-diffusion-xl",
            "prompt": prompt,
            "width": 1024,
            "height": 1024,
            "num_images": count,
            "steps": 30,
            "guidance_scale": 7.5
        }
    )
    
    result = response.json()
    return [img["url"] for img in result["images"]]

# Generate 4 variations
images = generate_batch_images(
    "A futuristic city skyline at night, neon lights, cyberpunk style"
)

Seed Control and Reproducibility

Using Seeds
Seed Variations
Random Seeds

Control randomness for reproducible results:

{
  "model": "stable-diffusion-xl",
  "prompt": "A red sports car on a mountain road",
  "seed": 123456789,
  "width": 1024,
  "height": 1024,
  "steps": 30,
  "guidance_scale": 7.5
}

Seed Benefits:

Reproduce exact same image
Create systematic variations
A/B test different parameters
Debug generation issues

Generate similar but different images:

base_seed = 123456789

for i in range(5):
    response = requests.post(api_url, json={
        "model": "stable-diffusion-xl",
        "prompt": "A forest path in autumn",
        "seed": base_seed + i,  # Slight seed variation
        "width": 1024,
        "height": 1024
    })

Let the system choose random seeds:

{
  "model": "stable-diffusion-xl", 
  "prompt": "Abstract art composition",
  "seed": null,  // or omit seed parameter
  "num_images": 4
}

The API will return the seeds used for each image for future reference.

Image Enhancement and Post-Processing

Upscaling and Super-Resolution

Real-ESRGAN Upscaling

Enhance image resolution and quality

curl

curl -X POST "https://api.comput3.ai/v1/enhance/upscale" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://your-image-url.com/image.jpg",
    "scale_factor": 4,
    "model": "realesrgan"
  }'

Options:

Scale factors: 2x, 4x, 8x
Models: RealESRGAN, ESRGAN, SRCNN
Cost: $0.05 per upscale operation

Face Enhancement

Improve facial details and quality

{
  "image_url": "https://your-image-url.com/portrait.jpg",
  "enhancement_type": "face",
  "strength": 0.8,
  "preserve_identity": true
}

Features:

Enhance facial features and skin texture
Preserve original identity and characteristics
Adjustable enhancement strength
Cost: $0.03 per enhancement

Background Removal

Remove or replace backgrounds

response = requests.post(
    "https://api.comput3.ai/v1/enhance/background",
    headers={"Authorization": "Bearer YOUR_API_KEY"},
    json={
        "image_url": "https://your-image.jpg",
        "action": "remove",  # or "replace"
        "new_background": "solid_white"  # if replacing
    }
)

Style Transfer and Artistic Effects

Neural Style Transfer
Preset Artistic Filters

Apply artistic styles to existing images

{
  "content_image": "https://your-photo.jpg",
  "style_image": "https://style-reference.jpg", 
  "strength": 0.7,
  "preserve_content": true
}

Quick artistic transformations

filters = [
    "oil_painting", "watercolor", "pencil_sketch", 
    "pop_art", "impressionist", "abstract"
]

for filter_name in filters:
    response = requests.post(api_url, json={
        "image_url": "https://your-image.jpg",
        "filter": filter_name,
        "intensity": 0.8
    })

Commercial Use and Licensing

Usage Rights and Licensing

Images generated on Comput3 Network come with different licensing options depending on your plan and intended use.

Personal Use License

Included with all plans✅ Allowed:

Personal projects and portfolios
Educational and research purposes
Social media posts (personal accounts)
Non-commercial art and creativity

❌ Not Allowed:

Commercial sales or licensing
Business marketing materials
Stock photo services
Resale or redistribution

Commercial Use License

Available with Pro and Enterprise plans✅ Allowed:

Business marketing and advertising
Product packaging and branding
Website and app content
Print materials and merchandise
Client work and services
Stock photo creation

Additional Requirements:

Attribution may be required for certain uses
Some models have specific commercial restrictions
Enterprise plans include full commercial rights

Extended Commercial License

Available as add-on for high-volume use✅ Includes:

Unlimited commercial usage
Resale and redistribution rights
No attribution requirements
White-label usage rights
Custom licensing terms available

Pricing: Contact sales for custom pricing

Model-Specific Considerations

Different AI models may have varying licensing restrictions. Always check the specific terms for the model you’re using.

Open Source Models
Proprietary Models

Stable Diffusion and derivatives

Generally permissive licensing
Commercial use typically allowed
May require attribution in some cases
Check specific model cards for details

Troubleshooting and Tips

Common Issues and Solutions

Poor Image Quality

Symptoms: Blurry, low-detail, or distorted imagesSolutions:

Increase the number of steps (30-50)
Adjust guidance scale (7-12 for most models)
Use negative prompts to exclude quality issues
Try different sampling methods
Increase resolution if budget allows

Negative prompt: blurry, low quality, distorted, bad anatomy, 
deformed, ugly, pixelated, grainy, artifacts

Prompt Not Followed

Symptoms: Generated image doesn’t match the descriptionSolutions:

Be more specific and detailed in prompts
Use attention weighting: (important detail:1.3)
Break complex prompts into simpler parts
Try different guidance scale values
Use negative prompts to exclude unwanted elements

Example Improvement: ❌ “A person in a room” ✅ “A young woman with brown hair sitting in a modern living room, natural lighting, realistic photography style”

Inconsistent Results

Symptoms: Wildly different outputs for same promptSolutions:

Use seed values for reproducible results
Increase guidance scale for more prompt adherence
Use more specific and detailed prompts
Try different sampling methods
Generate multiple images and select best results

Anatomical Issues

Symptoms: Incorrect hands, faces, or body proportionsSolutions:

Use negative prompts: bad anatomy, extra limbs, deformed hands
Try models specifically trained for human subjects
Use reference images or controlnets
Generate multiple versions and select best anatomy
Consider post-processing with face/hand enhancement

Optimization Tips

Cost Optimization

Start with lower resolution for iteration
Use faster models for experimentation
Batch similar requests together
Optimize prompts to reduce generation attempts

Quality Optimization

Spend time crafting detailed prompts
Use appropriate negative prompts
Choose the right model for your use case
Experiment with different parameters

Speed Optimization

Use lower step counts for drafts
Choose faster models when quality allows
Batch multiple images in single requests
Use appropriate resolution for final use

Workflow Optimization

Save successful prompts and settings
Use seeds for reproducible results
Create prompt templates for common use cases
Organize generated images with metadata

Next Steps

Video Generation

Learn how to create AI-generated videos and animations.

API Reference

Complete API documentation for image generation endpoints.

Advanced Techniques

Explore advanced techniques like ControlNet, LoRA, and custom models.

Getting Started

API

Chat

IDE/CLI

Launch GPU

Generate Medias

COM Token

MCP

ELIZAOS

​Available Image Generation Models

​Quick Start

ComfyUI Template

Recommended GPU

​Advanced Prompt Techniques

​Prompt Structure Best Practices

​Prompt Weighting and Control

​Model-Specific Guidelines

​Stable Diffusion XL (SDXL)

Strengths

Optimal Settings

​Stable Diffusion 2.1

Strengths

Optimal Settings

​DALL-E Style Model

Strengths

Optimal Settings

​Generation Parameters

​Resolution and Aspect Ratios

​Quality vs Speed Settings

​Batch Generation and Variations

​Generating Multiple Images

​Seed Control and Reproducibility

​Image Enhancement and Post-Processing

​Upscaling and Super-Resolution

​Style Transfer and Artistic Effects

​Commercial Use and Licensing

​Usage Rights and Licensing

​Model-Specific Considerations

​Troubleshooting and Tips

​Common Issues and Solutions

​Optimization Tips

Cost Optimization

Quality Optimization

Speed Optimization

Workflow Optimization

​Next Steps

Video Generation

API Reference

Advanced Techniques

Available Image Generation Models

Quick Start

Advanced Prompt Techniques

Prompt Structure Best Practices

Prompt Weighting and Control

Model-Specific Guidelines

Stable Diffusion XL (SDXL)

Stable Diffusion 2.1

DALL-E Style Model

Generation Parameters

Resolution and Aspect Ratios

Quality vs Speed Settings

Batch Generation and Variations

Generating Multiple Images

Seed Control and Reproducibility

Image Enhancement and Post-Processing

Upscaling and Super-Resolution

Style Transfer and Artistic Effects

Commercial Use and Licensing

Usage Rights and Licensing

Model-Specific Considerations

Troubleshooting and Tips

Common Issues and Solutions

Optimization Tips

Next Steps