Estimate Spark costs before production.

Spark is Sogni's stable render-credit unit. One Spark is pegged to a baseline SDXL 1024 fast render at 20 steps and maps to $0.005 of render value. Pack prices and SOGNI token conversion can differ from that benchmark.

Premium Spark is Spark from paid or premium sources, including in-app purchases, crypto purchases, worker earnings claimed as Spark, and contest winnings. Premium Spark unlocks premium model gates and is spent before free Spark. External API models such as GPT Image 2.0, Seedance 2.0 / Fast / Mini / 2.5, HappyHorse 1.1, Wan 3, and Wan 3 Uncensored need eligible purchased Premium Spark coverage; Spark bought with crypto counts toward it exactly like card-purchased Spark.

Premium licensed and external vendor models have extra licensing or vendor costs. Worker-backed premium jobs can support Spark or SOGNI where enabled. GPT Image 2.0, Seedance 2.0 / Mini / 2.5, HappyHorse 1.1, Wan 3, and Wan 3 Uncensored are stricter external API jobs: they require Premium Spark, do not offer SOGNI fallback, and are checked against purchased Premium Spark coverage (card or crypto).

At default settings: LTX-2.3 Distilled is about 17.5 Spark, LTX-2.3 Dev is about 43.8 Spark, WAN LightX2V is about 46.2 Spark, WAN Full is about 115.4 Spark, Seedance 2.0 Mini 720p is about 98.3 Spark, Seedance 2.0 720p is about 196.6 Spark, Seedance 2.5 720p is about 300.5 Spark (133.6 Spark at 480p), and HappyHorse 1.1 720p is about 182 Spark. Seedance V2V uses lower video-reference rates: about 59.0 Spark for Mini 720p and 120.7 Spark for Seedance 2.0 720p; Seedance 2.5 jobs with a video input are likewise cheaper per second than its text-to-video rate. Seedance 2.0 Mini is temporarily discounted through 7 September 2026 — see the promotion note above. Legacy Fast requests route to Mini. MiniMax H3 cannot render exactly 5 seconds - its shortest clip is 124 frames at 24fps, which is 5.17 seconds: 82.67 Spark on Standard, 51.67 on Balanced, 31 on LightX2V Turbo, or 20.67 on FastH3 at 544/768p-class output. An exact 8-second FastH3 clip is 192 frames and costs 32 Spark ($0.16).

MiniMax H3 is priced per second of finished output. FastH3 bills 4 Spark per second ($0.02) at both 480p and 544/768p-class output, making an exact 8-second clip 32 Spark ($0.16). At 480p, Standard, Balanced, and LightX2V Turbo bill 10, 6, and 4 Spark per second. At 544/768p-class output, they bill 16, 10, and 6 Spark per second. Aspect ratio and generation mode do not affect the output rate within a tier. Ref2VA bills exact reference-video input duration at the full resolution-tier rate: 10 Spark ($0.05) per second at 480p or 16 Spark ($0.08) at 544/768p-class, regardless of its Standard, Balanced, or LightX2V Turbo output rate. Standard and Balanced Ref2VA include five reference images and add 16 Spark ($0.08) per extra image. Reference audio is free, and FastH3 has no Ref2VA route. Two-stage IDs (FastH3 Two-Stage, and Two-Stage Ref2VA on Standard or Balanced) deliver 2× the requested canvas and add a per-second surcharge by canvas class on top of the tier's 544/768p-class rate — none at 768p, 6 Spark ($0.03) at 1080p, and 12 Spark ($0.06) at 2K; Two-Stage Ref2VA currently admits only 768p. H3 runs at fixed 24fps on the 124 + n×17 frame grid, so output clips run from 124 frames (5.17 s) to 362 frames (15.08 s).

On Sogni's primary NVIDIA RTX PRO 6000 H3 workers, a highest-quality 15-second standard H3 video typically takes about 15 minutes once rendering starts; LightX2V H3 Turbo typically takes about 5 minutes, and FastH3 Turbo — the default Turbo engine in Sogni Web, up to 6× faster than standard H3 on a 15-second clip — brings the same clip to roughly two minutes. Raw render time is the same on Unlimited and Unlimited Pro, but Pro adds another standard H3 concurrency slot, higher subscription queue priority, and twice the H3 fair-use capacity. Sogni does not publish a fixed 24-hour video count; the average paid subscriber does not hit daily throttling, and actual throughput varies with workflow, worker availability, queue demand, and adaptive fair-use controls. Sogni is testing community workflow optimizations that may improve H3 speed by up to 20%. Individual creators can keep substantial queues moving throughout the day through the API or Sogni Creative Agent Skill; unattended continuous 24/7 infrastructure and multi-user production workloads are not fair use.

No. Seedance is an external ByteDance/BytePlus vendor path. Sogni validates, prices, dispatches, and stores the result, but the render does not consume Sogni GPU worker capacity.

GPT Image 2 includes output-image token cost, estimated reference-image input token cost, and a 3,200 text-input token buffer when references are present. Reference inputs are estimated at 6,240 image-input tokens each. In multi-output referenced batches, the first output uses uncached image-input pricing and later outputs use cached image-input pricing.

The Sogni model id is gpt-image-2. Product copy may say GPT Image 2.0 or GPT Image 2, but the estimator and backend route both refer to the same external OpenAI image model.

Seedance video-reference and V2V jobs use the hasVideoInput pricing path. The estimate is based on generated duration, selected resolution, fixed 24fps output, and the current ByteDance/BytePlus rate table, not the source video's duration.

Free Boost Spark is intended for standard creation. Current backend rules allow free Spark API usage for Krea 2 Turbo and do not allow premium-only model gates such as Seedance 2.0 / Fast / Mini / 2.5, HappyHorse 1.1, or GPT Image 2.0. Collecting and spending it also requires a verified payment method — see the next question.

Free Spark was released for years behind signup-only checks, and a large amount of it ended up on accounts created purely to collect it. Verifying a payment method once makes that far more expensive than creating another email address, which keeps the free tier generous for people actually creating with it.

Either of these unlocks free Spark permanently: a 3-day free Unlimited trial (cancel any time — the Spark you have already earned stays yours), or any Premium Spark purchase by card or crypto. Accounts that have already shown a real payment identity never see the requirement: current and past subscribers, anyone who has ever held Premium Spark including Spark gifted by Sogni, anyone who has run a worker node, and testnet participants.

This applies to free Spark only. Premium Spark, SOGNI token, and Unlimited plan rendering are unaffected, and a locked account keeps its free Spark balance — it becomes spendable as soon as the account is verified. Free on-device rendering in the Mac and iOS apps is also unaffected, because it does not use the Supernet.

Yes, where the relaxed network is available. Relaxed network jobs use a 0.5x Spark multiplier compared with fast. Video, audio, LLM, and external vendor models are fast-only in the current implementation.

Some edit and context models use longer benchmark timings when reference images are attached. GPT Image 2.0 also charges estimated reference-image input tokens, plus a text-input buffer when references are present. Seedance video references switch to the video-input rate path. MiniMax H3 Ref2VA bills exact reference-video input duration at the full resolution-tier rate: 10 Spark ($0.05) per second at 480p or 16 Spark ($0.08) at 544/768p-class, even when output uses Balanced or LightX2V Turbo pricing. Standard and Balanced Ref2VA include five images and charge 16 Spark ($0.08) per additional image; reference audio is free. FastH3 has no Ref2VA route.

Spark packs are one-time purchases and do not require a monthly subscription. Failed paid jobs are normally credited automatically, while completed outputs are subject to expected model and seed variance. Purchase, refund, and generation-credit terms are covered by the Refund & Generation Policy.

Yes for supported worker-backed jobs. Spark is fixed-price; SOGNI conversion uses live market pricing with a fallback rate when market data is unavailable. GPT Image 2.0, Seedance 2.0 / Fast / Mini / 2.5, HappyHorse 1.1, Wan 3, and Wan 3 Uncensored external API jobs require Premium Spark and do not offer SOGNI fallback.

Yes, at dashboard.sogni.ai, for both Premium Spark top-ups and Unlimited plans. Accepted stablecoins are USDC on Base, Etherlink, and Ethereum, plus USDT on Ethereum. Sogni shows one receiving address and one exact amount per invoice; send that exact amount on the network shown, within the 60-minute window, and the credit posts automatically. Crypto plans are prepaid one-time purchases that do not auto-renew and have no free trial. See Pay with Crypto.

Sogni Unlimited is $20/month or $199/year (~17 % saving). Unlimited Pro is $50/month or $498/year. Annual plans are discounted in exchange for a one-year commitment. App Store and Google Play prices are shown by the platform in local currency and may differ from the Stripe web prices. The same prices can be prepaid in USDC or USDT — see Pay with Crypto. A 3-day free trial is available for new subscribers on card plans (one per person, payment method required); billing starts automatically after the trial unless you cancel. Cancel before the trial ends to avoid the first charge while keeping access until the original end date. The trial includes up to 120 images and 15 videos per day, including up to 6 MiniMax H3 videos (reset at midnight UTC), with trial video length limits; regular plan allowances begin once your paid period starts. Canceling renewal on web/Stripe, the App Store, or Google Play keeps trial access and its limits until the original trial end date, even if you cancel right away. You never need to cancel to start paying: on the web you can skip the trial at checkout, or end a running trial early from Wallet & Billing (same plan and cadence, charged today). Apple and Google Play trials cannot be converted to paid early; cancel renewal in the store to avoid the first charge. If you subscribe again after using a trial, checkout starts as a paid subscription.

Web/Stripe trials require a temporary card hold for the selected plan's payment amount, including the full annual amount for annual billing. You see the exact amount before checking an authorization box and selecting Accept & continue to Stripe, or you can choose Cancel trial setup to abandon setup without requesting a hold or starting a trial. Your bank may show SOGNI* NO CHARGE (wording may vary). Sogni never collects this verification hold or turns it into a payment. We request release immediately after the check; your bank may take a few business days to remove the pending item. Trial access starts after successful verification. We email you after the attempt with the amount, expected bank description, result, and release status. Reply to the email for help, including if your bank asks you to approve the authorization. See the card verification FAQ.

In Sogni Wallet, choose Update Plan, then Downgrade to Free. Review when your benefits end and what will no longer be included, then continue to Stripe to confirm. Apple and Google Play subscriptions must be canceled with the store that billed you. Paid access continues until the date shown by your billing provider; afterward, future generations use pay-as-you-go Spark. Your account, creations, and existing Spark balance remain available. See How to Cancel Sogni Unlimited (Downgrade to Free) for complete provider-specific steps.

Most supported open-source image, video, music, and LLM models are included on both tiers, including MiniMax H3 and H3 Turbo, Krea 2 Turbo, Dark Beast Krea 2 v3, Dark Beast Z-Image Turbo v9, One Obsession v22, Z-Image Turbo, Chroma, Qwen Image Edit 2511 Lightning, Wan 2.2, LTX-2.3, Ace-Step 1.5 XL music, and Sogni Intelligence LLM. GPT Image 2 / GPT Image 2.5 Sunburst / Flare, Seedance 2.0 / Fast / Mini / 2.5, HappyHorse 1.1, Wan 3, and Wan 3 Uncensored are not covered by any subscription tier — they always require Premium Spark. Unlimited and Unlimited Pro run the same models and differ by throughput, queue priority, and fair-use limits. Unless a release is specifically marked otherwise, newly supported open-weight models and related updates are automatically included for existing and new subscribers. See the full model table for details.

Unlimited subscription jobs use fair-use scheduling with active-job concurrency, queue-size limits, and unpublished adaptive daily throughput controls. Unlimited allows up to 4 image jobs and 1 video job in parallel, or 2 concurrent video jobs with fast/turbo models. Unlimited Pro allows up to 16 image jobs and 4 video jobs in parallel, with standard MiniMax H3 limited to 2. Queued media is capped at 64 total / 8 video on Unlimited and 192 total / 24 video on Pro. About 98% of normal fair-use subscribers never encounter daily throttling; exact thresholds are unpublished to prevent gaming and can shift with GPU supply and demand. Within each tier, daily usage is a scheduling signal so accounts with lighter use are favored. Sustained high-volume or automated use may be delayed or rate-limited. Premium Spark remains available for fastest paid priority.

Sogni is a decentralized worker network. Showing Fast and Relaxed capacity helps users see how much connected GPU time remained available for new jobs across completed UTC days. The active UTC day is omitted until it closes so partial readings do not distort the charts or their 30-day averages.

This page explains current product pricing logic and examples from the Sogni API and Socket code. Plan pricing, included features and models, usage allowances, fair-use controls, and other limits are subject to change at Sogni AI's discretion, subject to applicable law. Sogni will provide advance notice of material changes affecting an active paid subscription when required. Storefront taxes, currency conversion, promotional pricing, refunds, and future subscriptions may change the amount shown at checkout; the price and plan terms shown at checkout or in an applicable renewal notice are authoritative for that purchase or renewal. See the Cancellation and Refund Policy.