# AI Video Atlas: agent guide

AI Video Atlas (https://aivideoatlas.com) is a sourced directory of AI video: 107 real shots (prompt + model + estimated cost + creator credit), 181 workflows/tools/model guides, and vendor-stated prices per second for 12 model families. Data updated 2026-10-11.

## Which tool when
- User wants a whole video, ad or sequence planned → plan_shots(description, budget_usd?, style?, shots?). Follow its "steps" in order: DECIDE (models, modes, clip lengths, aspect, cost) → DIRECT (shot list + per-model prompts) → ASSETS (reference sheets, start/end frames, style ref; each with status needs-generation / user-supplied / approved) → IMAGE PROMPTS → **STOP: show the user every asset and get approval before generating ANY video** → generate shots → review. Global directives (same X in every shot, vertical, no text, style) apply to all shots; questions and request wording are ignored, not planned.
- Always adapt Atlas output to the user's voice, context and brand before showing it; keep prices, models, modes, clip lengths and reference tags exactly as returned.
- User describes a single shot or style → find_shot (filters: shot_type, model, max_cost_usd), then best_prompt(slug) for the full prompt.
- "Which model should I use / what will it cost?" → recommend_model(shot, budget_usd, clip_seconds, clips).
- "How much is X?" / "cheapest" → get_prices(model?, host?).
- "How do I do lip sync / consistent characters / upscale?" → get_workflow(job or query).
- "How do I prompt Kling / Veo / Seedance…?" → get_guide({model}). Topics: shot-vocabulary, tagging, consistent-characters, i2v-vs-r2v, setup-playbooks. plan_shots already returns a guide pointer per shot.
- "What's new?" → whats_new(days). Weekly digest for agents: https://aivideoatlas.com/digest/latest.json (also .md, archives at /digest/YYYY-MM-DD.json).

## Rules
- Generation modes (t2v, i2v, reference-to-video, start/end frame, v2v, lip sync) come only from vendor docs, with the source and check date; "unknown" modes are never recommended. plan_shots returns image prompts for any input frames, plus reference sheets when shots must match.
- Costs are estimates ("est.") from vendor-stated $/s; when a prompt states no clip length, 5 s is assumed. Legacy, deprecated and discontinued models are never used for estimates.
- "Not stated" / "unknown" means Atlas has no data. Never fill it in.
- Atlas does not judge quality. Rankings are by evidence (number of sourced shots) and price.
- Always credit the creator and link the source/Atlas URL when you reuse a prompt.
- Read-only and free; about 60 requests/min per client. On a 429, wait for Retry-After.

## Shot types
- ads: Ads that sell (21)
- music: Music videos (2)
- shorts: Viral shorts (23)
- characters: Consistent characters (26)
- product: Product shots (12)
- cinematic: Cinematic / drone (67)
- lipsync: Lip sync & dialogue (13)
- anime: Animation / anime (17)

## Jobs (get_workflow)
- consistent-character: Keep a character consistent
- lip-sync: Lip sync & talking heads
- product-ad: Product ads & UGC
- camera: Direct the camera
- long: Longer & multi-shot videos
- i2v: Animate a still image
- edit: Edit or restyle footage
- upscale: Upscale & smooth
- local: Run it on your own GPU
- models: Models & hosts

## Models
- seedance: Seedance, 40 shots, from $0.16/s
- kling: Kling, 24 shots, from $0.07/s
- veo: Veo, 20 shots, from $0.05/s
- luma: Luma, 17 shots, from $0.03/s
- runway: Runway, 13 shots, from $0.05/s
- hailuo: Hailuo, 12 shots, from $0.1/s
- sora: Sora, 6 shots
- ltx: LTX, 3 shots, from $0.06/s
- higgsfield: Higgsfield, 1 shots, from $0.0578/s
- wan: Wan, 1 shots, from $0.05/s
- grok: Grok Imagine, 0 shots, from $0.02/s
- vivix: Vivix, 0 shots, from $0.003/s