flux-2-klein
Generate Images with Flux 2 Klein on RunComfy
Also available from: doany-ai,agentspace-so
Image generation models can be difficult to prompt optimally. This skill provides Flux 2 Klein-specific patterns including step-count strategy, subject-first grammar, and model routing guidance for consistent professional results.
Download the skill ZIP
Upload in Claude
Go to Settings → Capabilities → Skills → Upload skill
Toggle on and start using
Test it
Using "flux-2-klein". A sleek wireless headphone product shot on matte white surface, soft studio lighting, 50mm lens, shallow depth of field, minimal background, professional product photography
Expected outcome:
A high-quality image of modern wireless headphones displayed cleanly on a white background with soft, professional studio lighting and subtle depth-of-field effect
Using "flux-2-klein". Same composition as reference, but change the background from outdoor park to indoor coffee shop with warm ambient lighting
Expected outcome:
The original subject and composition preserved, now set inside a cozy coffee shop with warm lighting, wooden furniture visible in the background, maintaining the original framing and subject pose
Using "flux-2-klein". A vintage botanical illustration of lavender in a copper vase, hand-drawn style, muted earth tones, textured paper background
Expected outcome:
An elegant vintage-style botanical rendering of purple lavender arranged in an antique copper vase, with delicate hand-drawn linework and a textured cream paper backdrop
Security Audit
Low RiskAfter evaluating the 83 static findings, 81 are false positives from markdown documentation being misidentified as executable code. The remaining findings reflect legitimate documented behavior: invoking the RunComfy CLI, network calls to model-api.runcomfy.net, and filesystem access for output files. Security documentation in SKILL.md (lines 217-223) confirms shell injection prevention, token file permissions (0600), and download URL whitelisting.
Medium Risk Issues (1)
Low Risk Issues (1)
Risk Factors
⚙️ External commands (4)
🌐 Network access (3)
📁 Filesystem access (1)
Quality Score
What You Can Build
Rapid Product Visualization
Iterate quickly on product designs with sub-second feedback from the 4B variant. Generate multiple concept variations in seconds for client reviews.
Brand-Consistent Asset Generation
Use multi-reference alignment to maintain consistent brand aesthetics across marketing materials, social posts, and campaign visuals.
Live Art Direction Sessions
Engage in real-time creative iteration with clients, adjusting prompts and styles instantly to explore creative directions together.
Try These Prompts
Generate a [product type] on a [surface/background], with [lighting style], [camera settings], professional e-commerce ready
A [subject] [action/pose] [setting details], [style/medium], [lighting], [camera/technical specs], [quality modifiers]
Same [composition/lighting/subject] as reference, but [specific changes]. Keep [unchanged elements] exactly as in reference.
Style consistent with reference images: [describe aesthetic elements present in all refs]. Subject: [new subject description]
Best Practices
- Front-load the subject in prompts: state what the subject is before describing actions or settings
- Use step counts strategically: 4-8 steps on 4B for rapid iteration, ~25 steps on 9B for final output
- Keep reference images stylistically consistent when using multi-reference workflows
Avoid
- Do not use conflicting adjectives like 'minimalist AND ornate' as they cancel each other out
- Do not exceed approximately 512 tokens in prompts as quality degrades without graceful truncation
- Do not request 4K resolution or ultra-wide aspect ratios beyond 16:9 as the model cannot produce them