Dataset·Catalogue

Generative Media Pipeline Patterns

Name: Generative Media Pipeline Patterns
Creator: Hermosa AI
Published: 2026-09-06
License: https://creativecommons.org/licenses/by/4.0/

Reusable multi-model chains: draft to finish, generate to edit, image to video, character-consistency rigs, 3D and audio-driven pipelines. The creative twin of the agent-patterns dataset, and a template library for the studio, each pattern is an ordered chain of model/node steps. Browse it below, or take the raw files.

Snapshot: 2026-09-06
Coverage: 77pipelines·9families·226chain steps
Download

About this dataset

[show]

Unit: One row = one reusable multi-model pipeline pattern, with its archetype family, input-to-output modality, why it needs more than one model, the ordered chain of model/node steps, the control levers, an example stack, use cases, the common pitfall, and a reference.
Steps as nodes: Each step carries a modelType drawn from the workflow node vocabulary (text-to-image, image-to-video, upscale, reference-adapter, image-to-3d, and so on), so a pattern maps directly onto workflow nodes and can seed a template.
Scope: Only genuine multi-model chains (two or more distinct generative models). A single unified model, a prompt trick, or a one-off workflow is out of scope.
Sources: arXiv papers, model documentation, and published ComfyUI / fal / Replicate workflows. References travel with the dataset.
License: Free to use, distribute, and reproduce with attribution to Hermosa Labs LLC (CC BY 4.0).
Cite: Hermosa Labs LLC, “Generative Media Pipeline Patterns,” 2026-09-06.

The catalogue

View

Family

77 of 77 pipelines

pipeline 1 of 77 · Draft to finish

Base + Refiner Draft-to-Finish

text -> image·SDXL base -> SDXL refiner -> 4x ESRGAN upscaler (or FLUX.1 schnell draft -> FLUX.1 dev refine).·Shared seed across passes; denoise strength on the refine step; base/refiner step-split ratio; optional control net on the draft.·One model rarely optimizes both global composition and fine detail at once; splitting draft from refine lets a cheap fast pass explore many options and an expensive pass finish only the keepers.·Over-denoising the refine step drifts away from the draft composition; a mismatched seed breaks continuity between passes.

Draft
SDXL baseFLUX.1 schnell
text-to-image
fast low-step base for composition and layout
Refine
SDXL refinerFLUX.1 dev
image-to-image
high-fidelity refiner pass for detail and artifact cleanup
Upscale
Real-ESRGAN 4x
upscale
optional 2-4x detail-preserving upscaler

Summary

A fast base model lays down composition and structure, then a high-fidelity refiner pass adds detail and fixes artifacts, optionally followed by an upscaler.

Stages: Draft: text-to-image (SDXL base, FLUX.1 schnell)
Refine: image-to-image (SDXL refiner, FLUX.1 dev)
Upscale: upscale (Real-ESRGAN 4x)
Use cases: Hero art and key visuals
Batch concepting then finishing keepers
Print-resolution stills

Generative Media Pipeline Patterns

Snapshot

2026-09-06

Coverage

77pipelines·9families·226chain steps

Download