Playcut AI models: specs, limits and credit pricing

Playcut gives you one API across models from Google, xAI, ByteDance and Alibaba — one API key, one credit balance, one async task shape. Switching from Veo 3.1 to Seedance 2.0 is a model field change, not an integration.

This page is the index: three image models, seven video models, two music models and three voice providers, with the numbers you actually need to pick one. Every model has its own page with per-mode detail and known failure modes.

How pricing works

Everything is billed in credits, held atomically at submission and refunded in full if the task fails or is cancelled. The reference rate used throughout these docs is $0.0145 per credit. Video models are priced per second of output; image and music models are priced per generation; TTS is priced per 1,000 characters of input text.

Image models

Three models, one order of magnitude apart in price. All three do text-to-image and reference-to-image (editing); only Nano Banana Pro does multi-region edit.

Model	Best for	Modes	Max resolution	Credits	Page
Nano Banana Pro gemini-3-pro-image-preview	Highest fidelity, text rendering, multi-region editing	text-to-image, reference-to-image, multi-region edit	4K	67 @ 1K/2K · 84 @ 4K · 25 flat per multi-region edit	Nano Banana Pro
Nano Banana 2 gemini-3.1-flash-image-preview	The workhorse — near-Pro quality at about half the price	text-to-image, reference-to-image	4K	34 @ 1K/2K · 43 @ 4K	Nano Banana 2
Grok Imagine (Image) grok-imagine-image	Cheap bulk generation, thumbnails, style exploration	text-to-image, reference-to-image	2K (no 4K)	10 text-to-image · 20 reference-to-image	Grok Imagine Image

Both Nano Banana models share the same ten aspect ratios (1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9). Grok Imagine Image is the only image model with an auto ratio and the only one offering 2:1, 1:2, 19.5:9, 9:19.5, 20:9 and 9:20 — but it stops at 2K.

Reference-image limits

Both Nano Banana models accept up to 14 reference images, matching Google’s documented composite limit (Pro: 6 objects + 5 characters + 3 style; Nano Banana 2: 10 objects + 4 characters). grok-imagine-image supports 2 (verified against xAI’s live API). Requests over a model’s cap are rejected before any credits are held. Details on each model page.

Video models

Seven models. They differ far more than the image models do — in modes, in maximum duration, and by 24× in price per second (10 credits/s on Seedance 2.0 at 480p up to 240 credits/s on Veo 3.1 at 4K).

Capabilities

Model	Best for	Modes	Aspect ratios	Max resolution	Duration
Veo 3.1 veo-3.1-generate-preview	Highest quality, native audio, 4K, the only model with all five modes	text-to-video, image-to-video, reference-to-video, interpolation, extension	16:9, 9:16 only	4K	4, 6 or 8 s (enum). Must be 8 at 1080p, 4K, with references, or for extension. Extension adds 7 s per call, up to 20 times
Grok Imagine (classic) grok-imagine-video	Cheapest video per second; widest ratio list	text-to-video, image-to-video, reference-to-video, extension	7 ratios (16:9, 4:3, 1:1, 9:16, 3:4, 3:2, 2:3)	720p	1–15 s text-to-video · 1–10 s for other modes · default 6 s
Grok Imagine 1.5 grok-imagine-video-1.5-preview	Better motion than classic, still cheap	image-to-video (text-to-video shipped by Playcut — see caveat)	inherited from classic, unverified	720p	1–15 s text-to-video · 1–10 s image-to-video
Seedance 2.0 seedance-2.0	Best capability-per-credit; the only non-Veo 4K	text-to-video, image-to-video, reference-to-video (+ audio ref), transition	21:9, 16:9, 4:3, 1:1, 3:4, 9:16	4K	4–15 s (integer), default 5 s
Gemini Omni Flash gemini-omni-flash-preview	Fixed-length 10 s animate clips from a still	image-to-video only on Playcut	output follows the source image	720p	Fixed 10 s (model actually returns 3–10 s; no duration parameter exists)
HappyHorse 1.1 happyhorse-1.1	Reference-heavy shots — up to 9 reference images; video edit with reference swaps	text-to-video, image-to-video, reference-to-video, video edit	all 9 the provider accepts	1080P	3–15 s (integer); edit follows the source (≤15 s)
Wan 2.7 wan-2.7	Lip-sync from driving audio, and the only video-edit model	image-to-video (audio-driven), reference-to-video, video-edit	16:9, 9:16, 1:1, 4:3, 3:4 — image-to-video takes no ratio at all	720p	4–15 s (video-edit: provider range 2–10 s)

Pricing and reference limits

Model	Credits per second	Max reference images	Page
Veo 3.1	160 @ 720p/1080p · 240 @ 4K	3 (reference) · 1 (animate) · 2 (interpolation) · 1 video (extension)	Veo 3.1
Grok Imagine (classic)	20 @ 480p/720p	7	Grok Imagine Video
Grok Imagine 1.5	32 @ 480p/720p	0 — 1.5 rejects reference_images with a 400	Grok Imagine Video
Seedance 2.0	10 @ 480p · 21 @ 720p · 60 @ 1080p · 108 @ 4K	7 on Playcut (provider allows 9)	Seedance 2.0
Gemini Omni Flash	40 @ 720p (400 credits for the fixed 10 s)	0	Gemini Omni Flash
HappyHorse 1.1	56 @ 720P · 72 @ 1080P	9 (reference) · exactly 1 first frame (animate)	HappyHorse 1.1
Wan 2.7	40 @ 720p	5 combined (reference) · 1 first frame (animate) · 1 source video (edit)	Wan 2.7

Two video caveats worth knowing before you spend credits

• Grok Imagine 1.5 text-to-video: Playcut accepts and bills it, but xAI’s model page lists 1.5 as image-to-video only. Prefer classic for text-to-video until this is confirmed.
• Seedance 2.0 and human faces: ByteDance rejects reference or input images containing real human faces with a privacy error — and the filter cannot reliably tell an AI-generated portrait from a photograph. Credits are held before dispatch, so a rejected submit is refunded, not free of round-trip time. Use Veo 3.1, HappyHorse 1.1 or Wan 2.7 for people.

Music models

Two Lyria 3 tiers from Google. Both are text-to-music and both accept up to 10 reference images as multimodal mood input — there is no audio-reference input.

Model	Best for	Output	Duration	Max references	Credits	Page
Lyria 3 Clip lyria-3-clip-preview	Loops, stings, background beds, fast iteration	44.1 kHz stereo MP3	Fixed 30 s — no duration parameter	10 images	20 per song	Lyria 3
Lyria 3 Pro lyria-3-pro-preview	Full arranged tracks with verse/chorus structure	44.1 kHz stereo, MP3	~2 minutes, influenced by the prompt only	10 images	40 per song	Lyria 3

Neither tier exposes a duration parameter. On Pro, arrangement is steered with [Verse] / [Chorus] / [Bridge] section markers in the prompt.

Voice models

Three providers behind one generate-tts tool. TTS is billed per 1,000 characters of input, rounded up, minimum one block.

Model	Provider	Best for	Provider input cap	Credits	Page
qwen3-tts-instruct-flash-2026-01-26	Qwen (default)	Everyday synthesis, lowest cost	600 characters	15 / 1,000 chars	TTS Voices
qwen3-tts-vd-2026-01-26	Qwen	Designing a new voice from a text description	not documented	20 flat per voice design	TTS Voices
qwen3-tts-vc-2026-01-22	Qwen	Cloning a voice from an audio sample	sample 5–60 s, ≤10 MB, ≥24 kHz, WAV/MP3/M4A	5 per clone	TTS Voices
gemini-3.1-flash-tts-preview	Gemini	Best stylistic control and language coverage	32k-token session context (Playcut DTO caps at 5,000 chars)	25 / 1,000 chars	TTS Voices

Qwen input length

Qwen voices are capped at 600 characters per request (the provider’s limit — enforced by Playcut before any credits are held). Use a Gemini voice for scripts up to 5,000 characters, or split longer text.

How to choose

Cheapest per output

• Image — Grok Imagine Image, 10 credits.
• Video — Seedance 2.0 at 480p, 10 credits per second. At 720p, Grok Imagine classic (20) and Seedance (21) are effectively tied.
• Voice — Qwen, 15 credits per 1,000 characters.
• Music — Lyria 3 Clip, 20 credits per 30-second song.

Highest quality

• Image — Nano Banana Pro, the only image model with multi-region editing.
• Video — Veo 3.1, the only model with native synchronized audio and the only one reaching 4K.
• Voice — Gemini, the priciest TTS we sell at 25 credits per 1,000 characters.

Longest single clip

• Seedance 2.0, HappyHorse 1.1 and Wan 2.7 all reach 15 seconds in one generation. Both Grok Imagine models also reach 15 seconds, but only on text-to-video — their image- and reference-driven modes cap at 10 seconds.
• Veo 3.1 caps a single clip at 8 seconds but is the only model that can extend — 7 seconds per call, up to 20 times.

Needs lip-sync from an audio track

• Wan 2.7 only. It is the sole model that takes a driving audioAssetId on image-to-video.

Needs to edit an existing video

• Wan 2.7 only. Every other model 400-rejects video-edit on Playcut.

Needs many reference images

• Video — HappyHorse 1.1 (9) then Seedance 2.0 (7) and Grok Imagine classic (7).
• Image — a Nano Banana model. Avoid Grok Imagine Image for multi-reference work; only the first reference is sent.

Needs vertical or square video

• Anything except Veo 3.1, which is 16:9 and 9:16 only. The widest video ratio list is Grok Imagine’s 7 (16:9, 4:3, 1:1, 9:16, 3:4, 3:2, 2:3); Seedance 2.0 and HappyHorse 1.1 offer 6 each and are the only video models with 21:9.

Needs 4K

• Image — Nano Banana Pro (84 credits) or Nano Banana 2 (43 credits).
• Video — Veo 3.1 only, at 240 credits per second.

Cross-model references

Aspect ratios by model Which ratio every image and video model accepts, side by side — and which combinations get rejected.

Resolutions & pricing Every resolution tier across the catalog with its exact credit cost, so you can compare cost per second and per image.

Actor engine tiers

The actor-act pipeline exposes four engine tiers instead of raw model ids. Each maps onto a video model above, and each adds a flat 15 credit scene charge:

Tier	Underlying model	Credits per second
fastest (default)	Grok Imagine classic, reference-to-video	7
fast	Grok Imagine 1.5, image-to-video	17
plus	HappyHorse 1.1, reference-to-video	30–38
pro	Gemini Omni Flash	35

The tier label is the only engine identifier on the wire — the underlying provider is an implementation detail and can change.

Next steps

• Quickstart — first generation in under 5 minutes
• API Reference — every tool and its schema
• Rate Limits — concurrency and throughput per plan