Skip to Content
Slloy Docs
Business, Pricing & Channels

AI Models & Effects Catalog

Specifications, pricing, pixel constraints, and readiness gates for Slloy's 5 image models, 12 video models, and 15 photo/video effects.

Slloy decouples public product SKUs from upstream provider model IDs through three registries: server/src/imageModels.ts, server/src/videoModels.ts, and server/src/effects.ts.

1. Image Models Catalog (5 Active Models)

Defined in server/src/imageModels.ts (DEFAULT_MODEL = "seedream"). All image prices are fixed per generated image and rounded to multiples of 10 credits.

Registry idDisplay labelProviderUpstream modelIdCredits / ImageCost (CNY)Aspect RatiosRef ImagesBadge / FeaturedNotes & Constraints
seedreamSeedream 5.0arkdoubao-seedream-5-0-pro-26062830¥0.60 (>2.61MP tier)1:1, 16:9, 9:16, 4:3, 3:4, 4:5, 2:3, 3:1Yeshot (#2)Migrated from 5.0 Lite (2026-11-16) before Lite EOL (2026-11-24). 2K direct output.
seedream-5-5Seedream 5.5arkdoubao-seedream-5-5-26102830¥0.60 (>2.61MP tier)1:1, 16:9, 9:16, 4:3, 3:4, 4:5, 2:3, 3:1Yesnew (#1)Latest flagship; strongest multi-subject consistency and text layout.
qwen-imageQwen-Imagedashscopeqwen-image-3.020¥0.2161:1, 16:9, 9:16, 3:4, 4:3Yesnew (#3)mode: "multimodal", watermarkParam: true. Best bilingual poster/logo text rendering.
qwen-image-proQwen-Image Prodashscopeqwen-image-3.0-pro30¥0.60 (2K tier)1:1, 16:9, 9:16, 3:4, 4:3Yespro (#4)mode: "multimodal", watermarkParam: true. Higher complex instruction adherence.
z-image-turboZ-Image Turbodashscopez-image-turbo10¥0.101:1, 16:9, 9:16, 3:4, 4:3, 4:5, 2:3No (false)lite (#5)maxPrompt: 800 chars, watermarkParam: false. Entry model (100 signup credits = 10 images).

Aspect Ratios & Pixel Dimensions (ARK_SIZES, QWEN_SIZES, ZIMAGE_SIZES)

/api/generate validates the requested aspectRatio in two layers: first against the global IMAGE_RATIOS whitelist, then against model.sizes (supportsRatio(model, ratio)). Unsupported ratios return HTTP 400 BAD_ASPECT_RATIO rather than silently falling back to 1:1.

  • Volcengine Ark 2K Pixel Constraints (ARK_SIZES):
    • Ark Seedream 5.0 Pro and 5.5 enforce a minimum total pixel count of >= 3,686,400 pixels (2K tier) and require both width and height to be exact multiples of 16.
    • All ARK_SIZES entries sit above the ARK_IMAGE_PIXEL_TIER = 2,610,000 (2.61MP) billing boundary (¥0.60/image):
      • 1:1: 2048x2048 (4.19MP)
      • 16:9: 2560x1440 (3.69MP)
      • 9:16: 1440x2560 (3.69MP)
      • 4:3: 2304x1728 (3.98MP)
      • 3:4: 1728x2304 (3.98MP)
      • 4:5: 1728x2160 (3.73MP, Instagram feed grid)
      • 2:3: 1568x2352 (3.69MP, Pinterest native)
      • 3:1: 3840x1280 (4.92MP, Substack / beehiiv newsletter header)
  • DashScope QWEN_SIZES (strict official enum; 4:5, 2:3, 3:1 are rejected):
    • 1:1: 1328*1328, 16:9: 1664*928, 9:16: 928*1664, 3:4: 1104*1472, 4:3: 1472*1104
  • DashScope ZIMAGE_SIZES (512x512 to 2048x2048 range):
    • 1:1: 1024*1024, 16:9: 1280*720, 9:16: 720*1280, 3:4: 864*1152, 4:3: 1152*864, 4:5: 1024*1280, 2:3: 1024*1536

2. Video Models Catalog (12 Active Models)

Defined in server/src/videoModels.ts (DEFAULT_MODEL = "fast"). Base prices (credits5s) are for 5 seconds at 720p (or 480p for 480p-only models), with explicit overrides in RES_PRICE_OVERRIDE so every resolution tier is a clean multiple of 10 credits. 10s and 15s scale linearly (credits5s * (duration / 5)).

Registry idDisplay labelProviderUpstream modelId480p (5s)720p (5s)1080p (5s)Max Dur (Sold)Capabilities & Billing
miniSeedance 2.0 Miniarkdoubao-seedance-2-0-mini-26061560100—15s¥0.0218 / 1k tok (billed). Default video effect engine.
fastSeedance 2.0 Fastarkdoubao-seedance-2-0-fast-26012890180—15s¥0.0356 / 1k tok (billed). Default workbench video model (featured: 7).
proSeedance 2.0 PROarkdoubao-seedance-2-0-26012815030060015s¥0.0570 / 1k tok (est). Only Seedance model with 1080p (featured: 4).
v25Seedance 2.5arkdoubao-seedance-2-5-260628180360—15s¥0.0750 / 1k tok (est). Flagship motion & audio-visual quality (featured: 1).
v25-flashSeedance 2.5 Flasharkdoubao-seedance-2-5-flash-261115100200—30s (15s UI)¥0.0500 / 1k tok (est). audio: true (featured: 2).
wan-t2vWan 2.7 T2Vdashscopewan2.7-t2v—15025015s¥0.60/s (720p), ¥1.00/s (1080p). audio: true, multiShot: true (featured: 8).
wan3Wan 3.0dashscopewan3.0-video8018032030s (15s UI)¥0.36/s (480p), ¥0.72/s (720p), ¥1.44/s (1080p). audio: true, multiShot: true (featured: 6).
wan-fastWan 2.2 Litedashscopewan2.2-t2v-plus30——5s (fixedDuration)¥0.14/s (480p). resStyle: "size", fixedDuration: true (omits duration param upstream).
wan-i2v-flashWan 2.2 I2V · Silentdashscopewan2.2-i2v-flash30——5s (fixedDuration)¥0.10/s (480p). resStyle: "tier", requiresFirstFrame: true, fixedDuration: true.
wan-i2vWan 2.6 I2V · Silentdashscopewan2.6-i2v-flash—406015s¥0.15/s (720p), ¥0.25/s (1080p). silent: true (forces audio: false), requiresFirstFrame: true.
kling4Kling 4.0klingkling-4.0—18027030s (15s UI)¥1.00/s (720p), ¥1.50/s (1080p). version: "kling-4.0" (featured: 3).
kling3Kling 3.0 · Silentklingkling-3.0—11014010s¥0.60/s (720p), ¥0.80/s (1080p). silent: true (forces sound: "off"), featured: 5.
kling3-i2vKling 3.0 I2V · Silentklingkling-3.0—11014010sSame pricing as kling3; resStyle: "tier", silent: true, requiresFirstFrame: true.

[!NOTE] Count verification: The MODELS array in server/src/videoModels.ts contains 13 entries covering 12 distinct video product configurations (kling3 and kling3-i2v share modelId: "kling-3.0" for text-to-video and image-to-video endpoints respectively).

Token Calibration Constants (TOK_PER_SEC)

For Volcengine Ark Seedance models, token consumption was calibrated from live production samples on 2026-09-26:

  • 480p: 10,500 tokens/sec (measured 10,128 tok/s + 3% safety margin)
  • 720p: 22,500 tokens/sec (measured 21,780 tok/s + 3% safety margin)
  • 1080p: 50,475 tokens/sec (measured 49,005 tok/s + 3% safety margin)
  • TOK_PER_FRAME = 0: Live verification proved that passing a 188KB first-frame image to doubao-seedance-2-0-mini-260615 produced the exact same 50,638 token count as pure text-to-video.

3. Photo & Video Effects (15 Active Effects)

Defined in server/src/effects.ts. Each effect combines a single user-uploaded reference photo with a curated prompt template.

Photo Effects (10 Image Effects — Seedream I2I)

Effect idCreditsAspect RatioAdjustableDescription
ai-action-figure203:4trueCollectible vinyl action figure in blister box with accessories
ai-polaroid101:1true1990s instant-film Polaroid frame with flash aesthetic
chibi-maker101:1trueTwo-head-tall cute chibi character illustration
photo-to-line-art101:1trueClean black ink line art on pure white background
pet-portrait203:4trueClassical aristocratic oil-painting pet portrait in gilt frame
caricature101:1trueFriendly hand-drawn magazine caricature
album-cover201:1trueBold graphic vinyl album cover art
lookbook203:4trueHigh-end editorial fashion magazine lookbook page
movie-poster203:4trueBlockbuster teal-and-orange cinematic movie poster
comic-page203:4trueThree-panel sequential comic book page

Video Effects (5 Video Effects — Seedance I2V mini 720p 5s = 100 Credits)

All video effects default to VIDEO_MODEL = process.env.VIDEO_EFFECT_MODEL || "mini" at 720p for 5s, dynamically priced via effectCost(effect) -> videoCost(mini, "720p", 5) = 100 credits:

Effect idCreditsModel (720p 5s)AdjustableDescription & Safety Rules
animate-old-photos100minifalseVintage photo gently blinks, smiles, and turns head while preserving texture
ai-hug100minifalseTwo subjects step toward each other and share a warm, natural hug
ai-kiss100minifalseTwo subjects lean in slowly and share a brief gentle kiss
muscle-transform100minitrueAthletic muscular transformation with strict clothing-preservation rules (no nudity/tearing)
earth-zoom-out100minifalseContinuous straight-axis pull-back from subject -> city block -> coastline -> blue Earth sphere (no moon mention to prevent foreground occlusion)

4. Model Readiness Gate & Channel Fault Attribution

Following the 2026-11-17 ModelNotOpen incident (gen #585, where an API key was present but doubao-seedream-5-5-261028 had not yet been activated in the Volcengine Ark console), Slloy enforces a strict two-stage readiness contract:

  1. Explicit ready: true Flag (imageModels.ts):
    • Having an API key (isAvailable()) does not mean a model is ready for public traffic.
    • isReady(model) requires model.ready === true. Unverified models are excluded from publicModels() and listForFrontend().
    • If a client requests a known model ID with ready !== true, resolveForRequest(rawId) returns notReady: true and /api/generate rejects the request before charging credits—never silently substituting another model.
  2. Channel Fault Attribution (server/src/channelFault.ts & channelFaultAlert.ts):
    • Upstream errors are classified into three FaultKind buckets:
      • config: ModelNotOpen, ModelNotFound, Arrearage / InsufficientBalance, InvalidApiKey, or HTTP 401/403. Never retried; immediately triggers reportConfigFault() to alert operators via structured error logs and Resend email.
      • transient: HTTP 429, HTTP 5xx, or pre-send connection errors (ECONNREFUSED, ENOTFOUND, EAI_AGAIN). Safe to retry (safeToRetry() === true) because no upstream artifact was billed.
      • permanent: All other 4xx errors (content safety, invalid parameters) and timeouts (AbortError/TimeoutError, which are never retried to avoid double-billing non-idempotent image generations).

On this page