Image and video models

This reference covers the image and video models exposed by default in Sundream, including video editing and upscaling models. Each price comes from the shared model catalog and the same estimator used by the generation controls.

Prices are in credits (cr). They are generation estimates, not dollar purchase prices. Your plan determines how many credits you receive; see Credits and plans.

Image prices are for one output with no reference images. Video rates are per second of output, before rounding; the example clip column shows the rounded credit estimate. Sundream rounds the total request estimate up to a whole credit, rather than rounding each second separately. Rates displayed here use up to four decimal places.

Availability and routing can change through workspace or administrator settings. The estimate shown before you generate is the one to check for your actual settings. The studio’s model browser shows your available models. Regenerating is a new request.

Image models

Seedream 4.5

Bytedance image model. Strengths: Multi-reference, Text rendering, 1K/2K/4K.

Provider: openrouter.

Seedream 5.0 Pro

Bytedance image model. Strengths: Reasoning, Multi-reference editing, 1K/2K.

Provider: openrouter.

Reference images add 0.3 cr per input image, before rounding the request total. The table excludes this surcharge.

Seedream 5.0 Lite

Bytedance image model. Strengths: Reasoning, Example-based editing, 2K/4K.

Provider: openrouter.

Nano Banana

Google image model. Strengths: Fast, Image editing, References.

Provider: openrouter.

Token-priced image generation; displayed cost is an estimate. Actual OpenRouter usage is recorded with the result.

FLUX schnell

Black Forest Labs image model. Strengths: Fastest, Low cost, Flat price.

Provider: replicate.

Flat per-output Replicate pricing; no edit.

FLUX.1 dev

Black Forest Labs image model. Strengths: High fidelity, Image-to-image, Flat price.

Provider: replicate.

Supports one image-to-image source via Replicate.

FLUX1.1 pro

Black Forest Labs image model. Strengths: Photorealism, Prompt accuracy, Image prompt.

Provider: replicate.

Single-output professional tier; optional Redux image prompt.

Recraft V3

Recraft image model. Strengths: Text rendering, Brand design, Flat price.

Provider: openrouter.

Imagen 4

Google image model. Strengths: Photorealism, High quality, 1K/2K.

Provider: replicate.

Imagen 4 Fast

Google image model. Strengths: Lower cost, Fast, Photorealism.

Provider: replicate.

Lower-cost Replicate Imagen 4 variant.

Pruna FLUX Fast

PrunaAI image model. Strengths: Low cost, Fast, FLUX quality.

Provider: replicate.

Replicate PrunaAI optimized FLUX — very low-cost fast text-to-image.

Pruna P-Image

PrunaAI image model. Strengths: Fast generation, Prompt adherence, Text rendering.

Provider: replicate.

Replicate PrunaAI P-Image — sub-second production text-to-image with strong prompt and text rendering.

Pruna Z-Image Turbo

PrunaAI image model. Strengths: Photorealism, Bilingual text, Low cost.

Provider: replicate.

Replicate PrunaAI-optimized Z-Image Turbo — fast photorealistic text-to-image with bilingual text rendering.

Luma Photon

Luma image model. Strengths: Cinematic, Lighting, Composition.

Provider: replicate.

Replicate Luma Photon — cinematic text-to-image.

GPT Image 2

OpenAI image model. Strengths: Instruction following, Text rendering, Precise editing.

Provider: openrouter.

Token-priced image generation at auto quality; displayed cost is an estimate. Actual OpenRouter usage is recorded with the result.

Video models

Each table lists the supported resolution tiers. Audio is included where supported unless a row says audio off. When an input type has a different price, it gets its own row. Reference-video rows use that route’s pricing; they do not imply that every resolution or input combination is accepted by the provider.

Wan 3.0

Wan 3.0 combines image, video, and audio references through Pika Omni. The character replacement flow uses your performance video, separate character portraits, and an approved appearance preview.

Provider: pika.

Inputs: start frame; end frame; up to 10 reference images; 1 reference video; up to 5 audio references; start frame + reference images: the frame is sent as a reference; start frame + a reference video: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or a reference video or audio references: not allowed.

Seedance 2.0 Fast

Bytedance's faster Seedance 2.0 tier, supporting text, image, video, and audio-conditioned video generation with native audio.

Provider: openrouter.

Inputs: start frame; end frame; up to 9 reference images; 1 reference video; up to 3 audio references; start frame + reference images: the frame is sent as a reference; start frame + a reference video: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or a reference video or audio references: not allowed.

Seedance 2.0

Bytedance's standard Seedance 2.0, supporting multimodal references, native audio, and 480p/720p/1080p/4K outputs.

Provider: openrouter.

Inputs: start frame; end frame; up to 9 reference images; 1 reference video; up to 3 audio references; start frame + reference images: the frame is sent as a reference; start frame + a reference video: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or a reference video or audio references: not allowed.

Seedance 2.0 Mini

Bytedance's lower-cost Seedance 2.0 Mini route, tuned for high-volume video with image/video references and native audio.

Provider: openrouter.

Inputs: start frame; end frame; up to 9 reference images; 1 reference video; up to 3 audio references; start frame + reference images: the frame stays exact; start frame + a reference video: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or a reference video or audio references: not allowed.

Seedance 2.5

ByteDance's Seedance 2.5 supports 4-30 second clips, multimodal references, native audio and dialogue at 480p/720p.

Provider: openrouter.

Inputs: start frame; end frame; up to 30 reference images; 1 reference video; up to 10 audio references; start frame + reference images: the frame is sent as a reference; start frame + a reference video: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or a reference video or audio references: not allowed.

Gemini Omni 1.1 Flash

Google's direct Gemini Omni 1.1 Flash route for text- and reference-led video, with native audio, first/last-frame interpolation, a low-cost 360p draft tier, and output up to 4K.

Provider: google.

Inputs: start frame; end frame; up to 3 reference images; no reference videos; no audio references; start frame + reference images: the frame stays exact; end frame + references: allowed.

Veo 3.1 Fast

Google's faster Veo 3.1 route with text/image-to-video, last-frame interpolation, 720p/1080p/4K output, and native audio.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

Veo 3.1 Lite

Google's cheapest Veo 3.1 route, with native audio at 720p/1080p. Takes text or start/end frames, but not a separate reference-image set.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

Veo 3.1

Google's flagship Veo 3.1 route for higher-fidelity text/image-to-video with synchronized audio and last-frame control.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

FLUX 3

Black Forest Labs' video model, generating up to 20 seconds with synchronized speech, effects, and ambience. Render time varies with settings and provider demand.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; 1 reference video; no audio references; start frame + a reference video: the frame is sent as a reference; end frame + a reference video: not allowed.

Runway Gen-4.5

Runway Gen-4.5 generates 2-10 second text- or image-led clips at 720p, with per-second pricing. Audio is added separately.

Provider: openrouter.

Inputs: start frame; no end frame; no reference images; no reference videos; no audio references.

Pruna P-Video

PrunaAI's p-video, an optimized text/image-to-video model tuned for speed and very low cost, with 720p/1080p output and optional last frame.

Provider: replicate.

Inputs: start frame; end frame; no reference images; no reference videos; 1 audio reference; start frame + audio references: the frame stays exact; end frame + references: allowed.

VEED Fabric 1.0

VEED Fabric turns a portrait plus a speech track into a talking video. One mode only: it lip-syncs supplied audio rather than generating a scene.

Provider: replicate.

Inputs: start frame; no end frame; no reference images; no reference videos; 1 audio reference; start frame + audio references: the frame stays exact.

Requires a portrait and a driving audio track. Output duration follows the audio; the example below uses 5 seconds. Audio generation is a separate request.

FLUX Video Upscale

Video-only 2x precise upscaling through OpenRouter. Available when OpenRouter video is enabled and this route is exposed.

Provider: openrouter.

Inputs: no start frame; no end frame; no reference images; 1 reference video; no audio references.

Requires a source video.

MiniMax H3

MiniMax H3 generates 5-15 second 2K video with native audio, landscape, portrait, and square formats, first/last-frame control, and up to three visual references.

Provider: openrouter.

Inputs: start frame; end frame; up to 3 reference images; no reference videos; up to 3 audio references; start frame + reference images: the frame is sent as a reference; start frame + audio references: the frame is sent as a reference; end frame + reference images or audio references: not allowed.

Runway Aleph 2.0

Instruction-guided editing of an existing source video. Available when OpenRouter video is enabled and this route is exposed.

Provider: openrouter.

Inputs: no start frame; no end frame; 1 reference image; 1 reference video; no audio references.

Requires a source video.

Minimum request estimate before rounding: 56 cr.

Grok Imagine Video 1.5

Grok Imagine Video 1.5 generates 1-15 second text- or image-led clips at 480p, 720p, or 1080p with native audio.

Provider: openrouter.

Inputs: start frame; no end frame; no reference images; no reference videos; no audio references.

Start-image requests include a fixed 1 cr surcharge. Their table rate is an effective average for the example duration; longer clips spread that fee across more seconds.

Grok Imagine

Grok Imagine generates 1-15 second clips at 480p or 720p, with support for up to seven reference images.

Provider: openrouter.

Inputs: start frame; no end frame; up to 7 reference images; no reference videos; no audio references; start frame + reference images: the frame is sent as a reference.

Start-image requests include a fixed 0.2 cr surcharge. Their table rate is an effective average for the example duration; longer clips spread that fee across more seconds.

Kling Video 3.0 Pro

Kling Video 3.0 Pro: cinematic text- and image-conditioned video from 3 to 15 seconds at 720p, with optional native audio and both start and end frame control.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

Kling Video 3.0 Standard

The economy tier of Kling Video 3.0 — the same 3 to 15 second range, 720p output, and audio toggle as Pro at a materially lower rate.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

Kling Video O1

Kling Video O1 generates 5 or 10 second clips at 720p, with optional native audio and start/end-frame control.

Provider: openrouter.

Inputs: start frame; end frame; no reference images; no reference videos; no audio references.

Wan 2.7

Wan 2.7 generates 2-10 second clips at 720p or 1080p, with optional native audio, start/end-frame control, and up to seven reference images.

Provider: openrouter.

Inputs: start frame; end frame; up to 7 reference images; no reference videos; no audio references; start frame + reference images: the frame is sent as a reference; end frame + reference images: not allowed.