Compose-director had contradictory instructions: Step 2 described Remotion captions/audio, but Steps 5/5b gave detailed FFmpeg code that agents followed instead. This caused three failures in production: FFmpeg subtitles instead of Remotion CaptionOverlay, missing audio (mixed externally but never embedded in Remotion props), and skipped audio verification in post-render review. Changes: - compose-director: Remotion is now DEFAULT for audio, captions, text overlays; FFmpeg is labeled FALLBACK only. Post-render review has mandatory ffprobe gate and audio transcription with explicit stop conditions. - remotion.md: routing table updated (captions/audio → Remotion), added universal Post-Render Verification Protocol for all pipelines (only 2/10 had one). - scene-director, asset-director: added pitfall for AI-generated text in CTA screens — must use Remotion text_card for verbatim text. - image-provider-usage: added Recraft V4 caveat (style param causes 422 on fal.ai). - recraft_image.py: documented the style parameter 422 issue inline.
6.1 KiB
Image Provider Usage for OpenMontage
How to choose between image generation and stock providers, and how to use each effectively. Supplements the existing
image-gen-usage.md(which covers FLUX prompting in depth).
Provider Landscape
Generation Providers (AI creates the image)
| Tool | Provider | Cost | Speed | Best For |
|---|---|---|---|---|
flux_image |
FLUX 2 Pro via fal.ai | ~$0.03-0.05 | ~5-10s | Photorealism, general purpose, workhorse |
grok_image |
Grok Imagine Image (xAI) | $0.02/output + $0.002/input edit image | ~5-15s | Image edits, style transfer, multi-image compositing |
openai_image |
GPT Image 1 (OpenAI) | ~$0.01-0.17 | ~5-15s | Complex instructions, text in images, multi-element |
recraft_image |
Recraft V4 via fal.ai | ~$0.04-0.25 | ~5-10s | Logos, SVG vectors, brand assets, text rendering (see caveat below) |
local_diffusion |
Stable Diffusion (local) | Free | ~30s+ | Offline, privacy, free |
image_gen |
Multi (legacy, deprecated) | Varies | Varies | Deprecated — use image_selector or per-provider tools |
Stock Providers (search and download existing images)
| Tool | Provider | Cost | Speed | Best For |
|---|---|---|---|---|
pexels_image |
Pexels | Free | ~2-5s | High-quality photography, color filtering |
pixabay_image |
Pixabay | Free | ~2-5s | Large library, category filtering, illustrations |
Selector
| Tool | Purpose |
|---|---|
image_selector |
Routes to the best available provider based on preference and availability |
Provider Selection by Scene Type
| Scene Type | Primary Provider | Why | Fallback |
|---|---|---|---|
| Real-world photo (city, nature, people) | pexels_image |
Real photos > AI for realism | pixabay_image → flux_image |
| Technical diagram | diagram_gen |
Structured, editable | flux_image with diagram prompt |
| Abstract/conceptual illustration | flux_image |
AI excels at custom concepts | openai_image |
| Style transfer / repaint of an existing image | grok_image |
Native edit flow, strong promptable transforms | openai_image |
| Multi-image merge / composite | grok_image |
Can combine multiple source images into one scene | openai_image |
| Logo or brand asset | recraft_image |
SVG support, text accuracy | openai_image |
| Image with text/labels | openai_image |
Best text rendering (GPT Image 1) | recraft_image |
| Complex multi-element composition | openai_image |
Best instruction following | flux_image |
| Hero image (key visual) | flux_image |
Highest visual quality | openai_image |
| Thumbnail | flux_image or recraft_image |
Needs to be eye-catching | — |
| Budget/free project | pexels_image or pixabay_image |
Free, immediate | local_diffusion |
| Offline/air-gapped | local_diffusion |
No network needed | — |
Provider-Specific Caveats
Recraft V4 via fal.ai
styleparameter causes 422 errors (as of 2026-04). Thestyleenum values (digital_illustration,realistic_image, etc.) are rejected by fal.ai's Recraft V4 endpoint. Workaround: encode style direction in the prompt text instead (e.g. "digital illustration of a tooth cross-section" rather thanstyle="digital_illustration"). Theimage_sizeandcolorsparameters work fine.- Text rendering is unreliable for exact business names. Recraft (like all AI image models) may hallucinate wrong text. For any scene where text must be verbatim (CTA screens, business names, phone numbers), use Remotion
text_cardinstead of generating an image with text.
Cost-Quality Tradeoff
PRODUCTION PATH: Premium
├── Hero images: flux_image ($0.05/img)
├── Supporting visuals: flux_image ($0.03/img)
├── Text overlays: openai_image ($0.04/img)
├── B-roll stills: pexels_image ($0.00)
└── Total for 10 images: ~$0.35
PRODUCTION PATH: Standard
├── All generated: flux_image ($0.03/img)
├── B-roll stills: pexels_image ($0.00)
└── Total for 10 images: ~$0.25
PRODUCTION PATH: Budget
├── All stock: pexels_image + pixabay_image ($0.00)
├── Diagrams: diagram_gen ($0.00)
└── Total: $0.00
PRODUCTION PATH: Offline
├── All generated: local_diffusion ($0.00)
├── Diagrams: diagram_gen ($0.00)
└── Total: $0.00 (but slower, lower quality)
Use generation_mode="edit" when the task starts from an existing image and should route only to edit-capable providers.
Using the Image Selector
For most cases, use image_selector and let it route:
# The selector finds the best available provider
result = image_selector.execute({
"prompt": "aerial view of a modern data center",
"preferred_provider": "auto", # or "flux", "pexels", etc.
"output_path": "assets/images/scene-3.png"
})
Override with preferred_provider when you know which provider is best for the scene type.
Use allowed_providers to restrict to free or local options:
# Budget mode: only free providers
result = image_selector.execute({
"prompt": "server room interior",
"allowed_providers": ["pexels", "pixabay", "local_diffusion"],
"output_path": "assets/images/scene-3.jpg"
})
Consistency Across Mixed Sources
When mixing stock and generated images in the same video, visual consistency is the challenge.
Strategy: Color Grade Everything
Apply the playbook's color grading LUT to both stock and generated images in the compose stage.
This unifies the look. The color_grade enhancement tool handles this.
Strategy: Match The Visual Identity, Not A Copied Prefix
When generating images, adapt the playbook's mood, palette, texture, and medium into a short scene-specific anchor. When searching stock, filter for the same emotional and visual qualities: color, lighting, environment, composition, era, and texture. Use the playbook as a consistency source, not a script to paste.
Strategy: Avoid Mixing Styles Within a Scene
Don't use a stock photo for one element and an AI illustration for another in the same scene. Keep each scene internally consistent — all stock or all generated.