Capability and input guide

Official channel: fast image creation and precise reference edits. High quality, PNG output, up to 16 references (4096px maximum per side). The displayed credit price includes your resolution and references. · Prompt: at most 4,000 characters. · image limit 20 MB · JPEG, PNG, WEBP · Masks, transparent backgrounds, custom sizes and additional quality tiers need dedicated controls. Output above 2560×1440 is experimental.

Official channel: detailed campaign images and controlled reference edits. High quality, PNG output, up to 16 references (4096px maximum per side). The displayed credit price includes your resolution and references. · Prompt: at most 4,000 characters. · image limit 20 MB · JPEG, PNG, WEBP · Masks, transparent backgrounds, custom sizes and additional quality tiers need dedicated controls. Output above 2560×1440 is experimental.

Fast 1K image generation and reference-led editing with up to ten images. · Prompt: at most 20,000 characters. · image limit 30 MB · JPEG, PNG, WEBP

Typography, page layouts and image editing with up to three references. One image per request at 1K or 2K.

Typography, page layouts and image editing with up to three references. One image per request at 1K or 2K.

V8.2 imagine/relax: four-image request (content review may return 1–4). Quality is serialized as a string. Existing runtime-verified V8.1 stays unchanged. Up to four references is Soho's tested product ceiling, not a claimed upstream maximum. · Fast and turbo have separate pricing; earlier versions, repeat, blend, edits and parent-task actions remain separate, unenabled workflows. · Character/style/depth references and weights, stylize, chaos, weird, tile, draft, seed and other native flags require their own validated controls.

1K/2K/4K, 16 documented ratios, and up to 16 public image references; 4K does not accept auto ratio. · The provider rejects automatic aspect selection at 4K. · Prompt: at most 32,000 characters.

1K/2K/4K, ten documented ratios, and up to 14 public image references. · Prompt: at most 32,000 characters.

Basic 1K or high 2K; up to ten references. Each reference after the first adds a provider charge. Layer and transparent modes remain gated. · Prompt: at most 4,000 characters. · Layer decomposition: variable 1–17 output billing is not mapped. · Transparent output: requires an alpha-bearing reference and a separate validated input mode.

0.5K/1K/2K/4K, 14 ratios, up to 14 references. Soho currently exposes one output per request; provider batches remain gated. · Prompt: at most 32,000 characters. · Batches of 2–4: provider-supported; multi-output quote/partial-settlement mapping pending. · Search grounding: server-owned and not enabled.

Approximately one-megapixel output, nine documented ratios, and up to four public references.

1–4MP output, nine ratio presets plus Soho's 4:5 pixel mapping, up to eight references; output plus references must total no more than 9MP.

Free-form width:height ratios; Soho's ten presets. Quality 0.25/0.5/1/2 on the pinned V8.1 relax route. Content review may return fewer than four images. · Prompt: at most 1,024 characters. · Fast/turbo, HD, repeat and follow-up actions require separate pricing and output contracts.

MixRoute documents 4–15 seconds, 480p/720p, seven ratios, first/last frames, and multimodal reference limits of 9 images, 3 videos, and 3 audio clips. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Soho request limit: 32,000 characters including the selected video look; not a provider prompt maximum. · Provider guidance: up to 1,000 English words or 500 Chinese characters is recommended, not required. Longer prompts are allowed within Soho's request limit.

MixRoute documents 4–15 seconds, 480p/720p, seven ratios, first/last frames, and multimodal reference limits of 9 images, 3 videos, and 3 audio clips. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Soho request limit: 32,000 characters including the selected video look; not a provider prompt maximum. · Provider guidance: up to 1,000 English words or 500 Chinese characters is recommended, not required. Longer prompts are allowed within Soho's request limit.

MixRoute documents 4–15 seconds, 480p through standard-only 4K, seven ratios, first/last frames, and multimodal reference limits of 9 images, 3 videos, and 3 audio clips. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Soho request limit: 32,000 characters including the selected video look; not a provider prompt maximum. · Provider guidance: up to 1,000 English words or 500 Chinese characters is recommended, not required. Longer prompts are allowed within Soho's request limit.

4–30 seconds, three resolutions, seven ratios, explicit frame roles, and up to 30 image references. Frame roles require adaptive ratio. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Frame inputs require adaptive aspect ratio. · Prompt including the selected video look: at most 20,000 characters. · image width/height 0.4–2.5 · image limit 30 MB · JPEG, PNG, WEBP, BMP, TIFF, GIF, HEIC, HEIF · Source video/audio and automatic duration are not enabled: they need input-duration pricing and validation.

Native audio always on. First/last frames derive ratio from input and reject aspect_ratio. Adaptive is reference/frame-only. References 6–9 incur extra image fees. Customer release remains gated. · A last frame needs a first frame. Frame inputs and general references are separate modes. · With frame inputs, aspect ratio follows the source image. · Prompt including the selected video look: at most 7,000 characters. · Image edges 256–5760 px · image width/height 0.4–2.5 · image limit 30 MB · JPEG, PNG, WEBP, HEIC, HEIF · Video/audio references require verified input duration and media limits.

Turbo route: 1–16 seconds, three resolutions; zero/one/two images select text/first/first+last. Frame modes reject aspect_ratio. Audio toggle does not change price. Customer release remains gated. · A last frame needs a first frame. Frame inputs and general references are separate modes. · With frame inputs, aspect ratio follows the source image. · Prompt including the selected video look: at most 2,000 characters.

2–30 seconds, five resolutions through 4K, adaptive plus five fixed ratios, frame mode or up to ten reference images. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Prompt including the selected video look: at most 20,000 characters. · Image edges 240–8000 px · image width/height 0.125–8 · image limit 20 MB · JPEG, PNG, BMP, WEBP · Source video/audio and automatic duration are not enabled: they need input-duration pricing and validation. · PNG references must be opaque (no alpha channel).

The same documented surface as Wan 3.0, serialized through Prime's dedicated frame and reference fields. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Prompt including the selected video look: at most 20,000 characters. · Source video/audio and automatic duration are not enabled: they need input-duration pricing and validation.

The Lite channel supports eight-second text-only clips at 720p/1080p/4K. No reference inputs or audio toggle; official/Fast/Quality are different routes. · Prompt including the selected video look: at most 4,000 characters.

Official channel: one required first frame, 6–15 seconds, 480p/720p, three ratios, and an audio toggle. · Prompt including the selected video look: at most 10,000 characters.

1–15 seconds, four resolutions, eight ratios, first-frame, paired first/last-frame, or up to seven fusion references. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Paired first/last frames: only 5 or 8 seconds. · Prompt including the selected video look: at most 5,000 characters.

3–15 seconds, std 720p/pro 1080p/4K. Three text ratios; frame mode follows the input ratio. Multi-shot and elements are separate gated modes. · A last frame needs a first frame. Frame inputs and general references are separate modes. · With frame inputs, aspect ratio follows the source image. · Prompt including the selected video look: at most 2,500 characters. · Multi-shot: up to five shots, 1–12 seconds and 500 prompt characters per shot; sum-of-shot-duration billing is not mapped. · Named elements: up to three elements with 2–4 JPG/PNG references each; separate input/ownership controls are not enabled.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.

Text or first/last-frame motion from 5–15 seconds at 480p or 768p. No general image references or audio toggle. · A last frame needs a first frame. Frame inputs and general references are separate modes. · With frame inputs, aspect ratio follows the source image.

Text, one first frame, one reference or three-image fusion. 4–10 second clips through 4K with flat per-generation pricing. · Prompt including the selected video look: at most 2,000 characters. · Source-video reference mode has separate pricing and requires its own owned-video workflow.

Text or a first frame with optional native sound, 5–20 seconds at HD or Full HD. · Multiple keyframes, source-video continuation and draft finalization need dedicated controls and pricing.

Text, one first frame OR 1–9 references. First frame: at least 300px per side, ratio 0.4–2.5. References: at least 400px per side. JPEG/PNG/WebP, at most 20MB each. Up to 5,000 non-CJK characters, or 2,500 when CJK text is present. No audio toggle or end frame. · With frame inputs, aspect ratio follows the source image. · Prompt including the selected video look: at most 5,000 characters (2,500 with CJK text). · Image edges at least 300 px · image limit 20 MB · JPEG, PNG, WEBP

Fast: text, first/last frames OR up to three reference images. Eight seconds only; per-generation pricing. No audio toggle. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Prompt including the selected video look: at most 4,000 characters. · GIF is documented only at 720p; disabled until GIF output storage is validated. · 15-second remix uses a separate parent-task endpoint and is not a 15-second generation option.

Official channel: text or first/last frames, 4/6/8 seconds, per-second pricing with separate audio tiers. No general references. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Prompt including the selected video look: at most 4,000 characters. · GIF is documented only at 720p; disabled until GIF output storage is validated.

Official channel: text or first/last frames, 4/6/8 seconds, per-second pricing with separate audio tiers. No general references. · A last frame needs a first frame. Frame inputs and general references are separate modes. · Prompt including the selected video look: at most 4,000 characters. · GIF is documented only at 720p; disabled until GIF output storage is validated.

Text, exactly one first-frame image OR exactly three fusion references. Two images are rejected; no first/last-frame mode. Duration-tier pricing is per generation, not per second. No documented audio toggle. · Prompt including the selected video look: at most 2,000 characters. · Source-video mode: one video at most 30s (first 10s used), optionally one image; duration must be omitted. Separate fixed-per-generation pricing. Disabled pending owned-video probe and source-driven output workflow.

Beta requires one JPEG/PNG/WebP image; 1–15 seconds, auto plus seven ratios. Native audio. Full reference documents content_filter, not the playground's nsfw_checker/audio controls; safety is always enabled. Official is cheaper but starts at six seconds. · Prompt including the selected video look: at most 4,096 characters. · JPEG, PNG, WEBP

Text OR one HTTPS first frame, 3–15 seconds. No end frame, general references, native audio or 4K. Aspect ratio is omitted when a first frame supplies framing. · With frame inputs, aspect ratio follows the source image. · Prompt including the selected video look: at most 3,072 characters. · Image edges at least 300 px · image width/height 0.4–2.5 · image limit 50 MB · JPEG, PNG

Requires 1–7 JPEG/PNG/WebP references plus one MP3 song lasting 10–300 seconds. Output follows the song; seconds are not a freely selected clip length. Resolution uses uppercase P. · Prompt including the selected video look: at most 3,000 characters. · Song length is measured from the owned MP3 before quoting and submitting. Subtitles, name, color and SRT overrides remain disabled pending subtitle-asset ownership and validation.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.

The model card reflects Soho's active contract. Broader provider features remain unavailable until separately accepted.