Partner Providers Runware

Runware Provider Management

Manage the lifecycle of Runware AI providers: discover new models from documentation and implement them as IMG.LY providers.

Key Files

• Provider Tracking: specs/providers/runware/providers.md
• Provider Specifications: specs/providers/ (schemas, architecture)
• Provider Implementations:
  • Image: packages/plugin-ai-image-generation-web/src/runware/
  • Video: packages/plugin-ai-video-generation-web/src/runware/
• API Patterns: specs/providers/runware/api-patterns.md
• README Documentation:
  • Image: packages/plugin-ai-image-generation-web/README.md
  • Video: packages/plugin-ai-video-generation-web/README.md

Workflow Overview

This skill operates in two phases:

1. Discovery (automatic): Check Runware docs for new/updated models
2. Implementation (user-approved): Create provider files for selected models

Phase 1: Discovery

Follow the checklist in DISCOVERY_CHECKLIST.md to:

1. Read current specs/providers/runware/providers.md to understand what's tracked
2. Fetch Runware provider documentation pages listed in the Provider Documentation Links table
3. Compare discovered models against the providers.md table
4. Report findings to user:
   • New models not in providers.md
   • Models with updated capabilities
   • Status recommendations (implement vs skip)

Discovery Output Format

Present findings as:

## Discovery Results

### New Models Found
| Provider | Model | AIR | Capabilities | Recommendation |
|----------|-------|-----|--------------|----------------|
| ... | ... | ... | ... | implement/skip |

### Updated Models
| Provider | Model | Change |
|----------|-------|--------|
| ... | ... | ... |

### Summary
- X new models found
- Y models updated
- Z recommended for implementation

After presenting results, ask: "Which models would you like me to implement?"

Phase 2: Implementation

Only proceed after user approval. Follow IMPLEMENTATION_CHECKLIST.md to:

1. Determine provider type (t2i, i2i, t2v, i2v) from capabilities
2. Find existing Runware provider of same type as reference
3. MANDATORY: Fetch and verify API documentation (see API Verification below)
4. For I2I/I2V: Extract dimension constraints (min/max/multiple) from API docs
5. Create provider implementation following patterns in specs/providers/schemas/
   • For I2I: Use model-specific RUNWARE_DIMENSION_CONSTRAINTS (see implementation-notes.md)
6. Export provider from runware/index.ts
7. Add translations to translations.json - REQUIRED
8. Add to example app (examples/ai/src/runwareProviders.ts) - REQUIRED
9. Update specs/providers/runware/providers.md status to "implemented"
10. Update README documentation - Add to plugin's README.md (Providers, API Reference, Panel IDs, Asset History)
11. Run build checks: pnpm --filter "@imgly/plugin-ai-*" check:all

API Verification (MANDATORY)

CRITICAL: Before writing ANY provider code, you MUST fetch and verify the exact API parameters from TWO documentation sources:

1. General API Reference - For the inference type (image or video):

   • Image: https://runware.ai/docs/en/image-inference/api-reference.md
   • Video: https://runware.ai/docs/en/video-inference/api-reference.md

   Extract: Allowed top-level parameters, request/response format, delivery methods

2. Provider-Specific Documentation - For model-specific settings:

   • https://runware.ai/docs/en/providers/{vendor}.md (e.g., google.md, bfl.md)

   Extract: providerSettings.{vendor} options, model-specific parameters, constraints

Verification Checklist:

• Fetched general API reference for inference type
• Listed all allowed top-level parameters from API docs
• Fetched provider-specific documentation
• Identified any providerSettings.{vendor} options (e.g., providerSettings.google.generateAudio)
• Verified parameter names match EXACTLY (no assumptions!)
• Noted any async/polling requirements (video uses deliveryMethod: "async")

Why This Matters: Provider-specific features (like audio generation) are often nested under providerSettings.{vendor} rather than being top-level parameters. Using wrong parameter names causes API errors.

Implementation Reference

Based on model capabilities, follow the appropriate schema documentation:

Status Values

When updating providers.md:

Documentation Access

Runware documentation URLs can be accessed as LLM-friendly markdown by appending .md to the path:

• HTML: https://runware.ai/docs/en/image-inference/api-reference
• Markdown: https://runware.ai/docs/en/image-inference/api-reference.md

Always prefer the .md version when fetching documentation for better parsing.

References

Skill Files (this folder)

• DISCOVERY_CHECKLIST.md - Step-by-step discovery process
• IMPLEMENTATION_CHECKLIST.md - Step-by-step implementation process

Provider Specifications

• specs/providers/README.md - Overview of provider system
• specs/providers/architecture.md - How providers fit into the plugin system
• specs/providers/patterns/ui-guidelines.md - CRITICAL: Which parameters to expose/hide in UI
• specs/providers/patterns/text-to-image.md - T2I implementation pattern
• specs/providers/patterns/image-to-image.md - I2I implementation pattern
• specs/providers/patterns/text-to-video.md - T2V implementation pattern
• specs/providers/patterns/image-to-video.md - I2V implementation pattern
• specs/providers/patterns/quick-actions.md - Quick action support

Runware-Specific

• specs/providers/runware/providers.md - Model inventory and status
• specs/providers/runware/api-patterns.md - Runware API conventions
• specs/providers/runware/implementation-notes.md - Runware-specific details