- Home
- Skills
- Image & Video
- Stable Diffusion Image Pipelines
More screenshots
Works with the AI Agents you already use
Stable Diffusion Image Pipelines
Build repeatable local image workflows with a draft-to-final ladder, scheduler and VRAM choices, controlled generation, candidate scoring, and a run record for every accepted image.
$5
Stable Diffusion Image Pipelines
Example session with this skill installed
Create a 16:9 website hero image: a small ferry approaching a rainy harbor at dawn, cinematic photograph, no visible brand marks and no people close to the camera. The workstation has 12 GB of VRAM. The image must be reproducible next month with the option to change the ferry color.
- Read your context and instructions
- Compiled the stable diffusion image
Brief translated into visible decisions: ferry placed left of center with clear water on the right for page copy; dawn light from behind the harbor; rain visible in the light, not as white streaks across the whole frame; one photographic medium; no close figures, logos, or readable signs.
Prompt: “a small passenger ferry approaching a rainy harbor at dawn, viewed from the water, low cinematic camera, wet decks and warm cabin windows, layered blue-gray rain, soft golden sunrise behind cranes and warehouses, photographic depth, restrained contrast.” Negative layers: permanent defects (duplicated boats, warped hull, malformed masts, watermark); photographic defects (illustration, plastic water, oversaturated sky); brief exclusions (close people, readable signage, brand marks, sunny weather).
Ladder: drafts at 768×432 with a distilled setup, six steps, three seeds. Seed 14 won on hull silhouette and copy space. Candidates used the full model, a multistep DPM-style scheduler, 28 steps, guidance 7.4, 1344×768, fixed seed. Two branches changed only guidance and adapter-free surface detail; the 7.4 branch kept the rain coherent without an overcooked sky. Final refinement used image-to-image at 0.32 strength on the winner.
Inspection: at page size, the ferry reads first and the right side remains calm enough for copy. At full resolution, one distant crane repeats a support, outside the focal area and disclosed in the selection note. Run record preserves the model hash, both prompts, seed, scheduler, steps, guidance, dimensions, memory options, and refinement strength. To change only the ferry color next month, rerun the recorded candidate settings with the color phrase replaced and keep the seed and composition controls fixed.
Connects securely to your tools. The creator never sees your data.
About this skill
A single lucky generation is not a pipeline. This skill turns local diffusion work into a repeatable production process: define the deliverable, write prompts in a deliberate order, build negative prompts by layer, choose the scheduler from the step budget, and plan VRAM before the first full render. You will climb a draft, candidate, and final ladder instead of paying full-render cost to discover a weak composition. The workflow covers text-to-image, image-to-image refinement, inpainting, structural controls, style adapters, identity consistency, hardware probes, and delivery-size inspection. It also gives you a run-record template so a winning image can be rerun with one controlled change, a six-dimension candidate rubric that rejects structurally broken favorites, and fixes for common failures such as plastic skin, prompt neglect, malformed interactions, corrupted text, seam defects, and out-of-memory errors during decode. The result is fewer random rolls, clearer selection decisions, and image sets that can be reproduced when a client asks for the same scene with one element changed.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
- Settings must be adapted to the model family and interface; a recipe for one checkpoint may not transfer directly to another.
- It cannot guarantee exact lettering inside generated imagery; important text is usually better added during design production.
- Local generation speed and maximum resolution depend on your hardware.
- The workflow supports rights-aware production but cannot determine whether a particular asset or likeness is cleared for your use.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago
Permissions
No special permissions declared or detected