Fal Implemented Models

Track implemented fal.ai models and compare against available endpoints.

Database File

Location: ../../models.json

JSON database of all fal.ai model endpoints with implementation tracking.

Schema

json

{
  "endpoint_id": "fal-ai/flux-1/dev",
  "category": "text-to-image",
  "display_name": "FLUX.1 [dev]",
  "status": "active",
  "implemented": true,
  "direct": false
}

•implemented: We have an inference.sh app for this endpoint
•direct: We integrate with the provider directly (google, xai, bytedance) rather than via fal.ai

Usage

Check if Model is Implemented

bash

jq '.[] | select(.endpoint_id == "fal-ai/flux-1/dev") | .implemented' models.json

List All Implemented Models

bash

jq '[.[] | select(.implemented)] | .[].endpoint_id' models.json

List Unimplemented Models

bash

jq '[.[] | select(.implemented == false)] | .[].endpoint_id' models.json

Filter by Category

bash

# All text-to-image models
jq '[.[] | select(.category == "text-to-image")]' models.json

# Unimplemented video models
jq '[.[] | select((.category | test("video")) and .implemented == false)]' models.json

Direct Integrations

bash

# List endpoints we integrate directly (not via fal.ai)
jq '[.[] | select(.direct)] | .[].endpoint_id' models.json

# Unimplemented direct integration candidates
jq '[.[] | select(.direct and .implemented == false)]' models.json

Get Stats

bash

jq '{
  total: length,
  implemented: [.[] | select(.implemented)] | length,
  by_category: (group_by(.category) | map({(.[0].category): length}) | add)
}' models.json

Mark Model as Implemented

bash

jq '(.[] | select(.endpoint_id == "fal-ai/new-model")).implemented = true' models.json > tmp.json && mv tmp.json models.json

Mark Multiple Models (by prefix)

bash

jq '[.[] | if .endpoint_id | startswith("fal-ai/wan-pro") then .implemented = true else . end]' models.json > tmp.json && mv tmp.json models.json

Updating the Database

To refresh from fal.ai API:

bash

# Fetch all pages from API
curl -s "https://api.fal.ai/v1/models?limit=100" > page1.json
# ... continue pagination until has_more is false

# Merge preserving implemented status
python3 << 'EOF'
import json

# Load existing
with open('models.json') as f:
    existing = {m['endpoint_id']: m['implemented'] for m in json.load(f)}

# Load new from API
# ... process API data ...

# Preserve implemented status
for m in new_models:
    m['implemented'] = existing.get(m['endpoint_id'], False)
EOF

Consolidation Strategy

fal.ai often has multiple functions as separate endpoints:

•fal-ai/model/text-to-image
•fal-ai/model/image-to-image

Our approach: Consolidate into single apps where practical:

•If input has image -> run image-to-image variant
•If input is text only -> run text-to-image variant
•Language variants (like TTS with multiple languages) -> single app with language dropdown

This keeps our ecosystem clean and reduces redundancy.

Example Workflows

"Do we have flux implemented?"

bash

jq '[.[] | select(.endpoint_id | test("flux")) | {endpoint_id, implemented}]' models.json

"What video models are we missing?"

bash

jq '[.[] | select((.category | test("video")) and .implemented == false)] | .[].endpoint_id' models.json

"Coverage by category"

bash

jq 'group_by(.category) | map({
  category: .[0].category,
  total: length,
  implemented: [.[] | select(.implemented)] | length
}) | sort_by(-.total)' models.json