Leonardo.Ai vs Elyumfor Wan 3.0
Elyum charges only on keep (15 credits / $0.15 for 5s at 480p). Leonardo.Ai deducts tokens on generate and does not publish Wan 3.0's cost.
Wan 3.0 is Alibaba's video model, in public beta since August 6, 2026. It is built for one-pass clips up to 30 seconds at 1080p, with speech, foley, and ambience generated in the same render as the picture. Reference control is the other half of the product: up to 10 images and 5 clips in one generate, plus an edit path that changes visuals, plot, or dialogue on a clip you already have. That is the 1080p 30-second ad job. You lock a face, a product, a location, and existing footage, then describe how they combine instead of stitching 15-second fragments.
The weights do not change between studios. The invoice does. Elyum publishes Wan 3.0 per second on the Wan 3.0 page (1 credit = $0.01) and charges only on keep. Leonardo.Ai sells Fast Tokens. Leonardo.Ai's pricing page lists Essential at $12/mo (8,500 fast tokens plus a 25,500-token rollover bank), Premium at $30/mo, and Ultimate at $60/mo. Wan 3.0's token cost was not independently verified. Failed generates should auto-restore tokens. Restoration is not necessarily instantaneous. 2025-2026 user reports describe tokens staying deducted until you email [email protected] with the generation ID and timestamp.
Leonardo.Ai is the better desk if you want canvas tools, team seats (Team Starter starts at $72/mo for three seats), and a rollover token bank next to Lucid Origin and Phoenix. Elyum does not have a timeline editor, team seats, a community feed, stock footage, avatars, or a mobile app. We only charge kept takes. If your Wan 3.0 clips already sit in a Leonardo.Ai project beside those stills, moving one model is extra friction.
Wan 3.0 on Leonardo.Ai
Leonardo.Ai offers Wan 3.0 at up to 30 seconds with native synced audio from text, frame, or mixed references, per its own model marketing copy. Leonardo.Ai does not publish the token cost per Wan 3.0 generate. Third-party video models consume tokens at full rate even on Ultimate ($60/mo). Relaxed, token-free generation on Premium and Ultimate is limited to first-party models (Lucid Origin, Phoenix, FLUX, Motion), so it does not cover Wan 3.0. Max resolution for Wan 3.0 on Leonardo.Ai was not independently verified. The Leonardo.Ai alternative page covers the rest of that catalog.
Support states tokens are deducted on generate and should restore automatically if a generation fails, but restoration is not necessarily instantaneous. Adobe Community and Trustpilot reports from 2025-2026 describe tokens consumed on failed or off-instruction jobs. The refund path is emailing [email protected] with the generation ID and timestamp. One Adobe Community report described an 800-credit loss after a video failed to generate. Free-tier web generations (150 tokens/day, no rollover) carry a Leonardo watermark and are public by default. Paid subscribers get private generations, no watermark, and full IP ownership.
Wan 3.0 on Elyum
Wan 3.0 on Elyum is 3 credits/second at 480p (5s = 15 credits = $0.15; 15s = 45 credits = $0.45), 5 credits/second at 720p (5s = 25 credits = $0.25; 15s = 75 credits = $0.75), and 10 credits/second at 1080p (5s = 50 credits = $0.50; 15s = 150 credits = $1.50). Clips are 2-30 seconds with native audio. Modes: text to video, image to video, reference to video, and video editing. Elyum pricing uses 1 credit = $0.01. Listed rates already sit about 20% under comparable platform anchors.
Every render returns as a free preview (a take). Credits charge only when you KEEP it. KILL refunds the rest in full. A failed generation is never charged. The Free plan includes up to 150 credits once (our algorithm may grant them), no card, and meters kills at one per month. Plus ($9/mo), Pro ($29/mo), and Ultra ($69/mo) kill freely. Pro and Ultra include fair-use unlimited image generation on selected image models, which does not apply to Wan 3.0. Kept downloads are clean and watermark-free, with commercial use included. You can call the same model from Claude, ChatGPT, or Cursor via the Elyum MCP server or the REST API.
What a Wan 3.0 take costs
| Scenario | Leonardo.Ai | Elyum |
|---|---|---|
| One 5s take you keep | Not published (Wan 3.0 token cost not independently verified) | 480p: 5 × 2.6 = 15 credits ($0.15); 1080p: 5 × 10.4 = 50 credits ($0.50) |
| Ten 5s 480p takes, keep two | Tokens charged on each generate; no keep/kill refund for unused takes | 2 × 15 credits = 30 credits ($0.30); eight kills refund in full |
| A failed generation | Tokens deducted, then auto-restore if it fails (not always instant); else email [email protected] | Never charged |
| Monthly plan needed to access the model | Not published; Fast Token plans start at Essential $12/mo. Free: 150 tokens/day. | No subscription required. Free: up to 150 credits once (our algorithm may grant them). Previews are free. |
| Watermark on the file | Free web generations: yes. Paid: no. | No watermark on kept downloads |
| One 15s 1080p keep | Not published | 15 × 10.4 = 150 credits ($1.50) |
Elyum: 1 credit = $0.01, charged only when you keep a take. Leonardo.Ai figures from its own pricing pages and independent reviews, verified 2026-09-07; sources at the end. Full rate card on the Wan 3.0 page.
Where the two differ
Elyum sells credits at $0.01 and lists Wan 3.0 per second. Leonardo.Ai sells Fast Tokens on the web app and a separate dollar API. Wan 3.0's token cost on Leonardo.Ai is not published. Trustpilot reviewers call that token math confusing once third-party video models enter the mix.
On Elyum, kill a preview for a full automatic refund; failures never bill. On Leonardo.Ai, tokens deduct on generate. Support says failed jobs should auto-restore, not necessarily instantly. User reports from 2025-2026 describe tokens staying gone until you email [email protected] with the generation ID and timestamp.
Both market Wan 3.0 up to 30 seconds with native audio. Elyum duration is 2-30 seconds at 480p, 720p, or 1080p. Leonardo.Ai's max resolution for Wan 3.0 was not independently verified.
Elyum runs one credit balance from the web studio and from Claude, ChatGPT, and Cursor via MCP, plus a REST API. Leonardo.Ai's API is a separate product: developers buy dedicated API credits and a bearer-token key. API usage is billed in dollars, not Fast Tokens.
Leonardo.Ai ships canvas and editing tools, team workspaces (Team Starter $72/mo, three-seat minimum), and a rollover token bank on paid plans. Elyum does not have a timeline editor, team seats, a community feed, stock footage, avatars, or a mobile app.
If you also run Kling, compare Leonardo.Ai vs Elyum for Kling 3. Seedance sits on the Seedance 2.5 page.
Pick Leonardo.Ai when
- Canvas, first-party image models (Lucid Origin, Phoenix, FLUX, Motion), and Wan 3.0 live in one Leonardo.Ai project.
- Paid plans include a rollover token bank (Essential: 8,500/mo plus a 25,500 bank) so unused tokens are not fully lost.
- Team Starter and Team Growth add shared seats and a shared token pool, which Elyum does not offer.
- Trustpilot reviewers frequently praise fast Leonardo.Ai support, including refunds when you send a generation ID.
Pick Elyum when
- Published Wan 3.0 rates: 2.6 / 5.2 / 10 credits per second at 480p / 720p / 1080p.
- Previews are free. You pay only for keeps. Kills refund in full. Failures never bill.
- Free plan is up to 150 credits once (our algorithm may grant them), no card. Plus starts at $9/mo, below Leonardo.Ai Essential at $12/mo.
- Same model from the browser or via MCP and the REST API, one credit balance.
Wan 3.0 in three styles, rendered on Elyum
Getting the most out of Wan 3.0
- Write a 30-second clip as three beats (setup, hold, payoff) with a clock time on each beat. Wan 3.0 generates the duration in one pass. If you only describe the opening action, the last 15 seconds often idle or morph.
- Put dialogue, room tone, and foley in the same prompt as the picture. Audio is native, not a second pass. Quote the line you want spoken. Name the room (HVAC, street bleed, dry interview booth) or the model invents a bed.
- When you attach references, label each one's job in text: image 1 face lock, image 2 product hero, images 3-4 wardrobe, clip 1 location plate. The cap is 10 images and 5 clips. Unlabeled stacks get averaged into a generic look.
- Give one camera move per beat (slow push-in, then a locked wide). Stacking dolly plus orbit plus whip-pan in one sentence fights the motion prior and smears faces over a 20-30s take.
- For edits, state the change only: swap the line at 0:08, recolor the can, keep blocking. Wan 3.0 can edit an existing clip. Regenerating the whole 30s from a new text prompt throws away the take you already liked.
- Lock resolution to the delivery: 480p for board reviews, 1080p for the keep. Prompt the same. Frame in a sentence (16:9 product plate, 9:16 vertical end card), not a pile of quality adjectives.
Frequently asked questions
How much does a Wan 3.0 video cost on Elyum vs Leonardo.Ai?
Is Wan 3.0 the same model on both?
What happens to a bad Wan 3.0 take?
Do I need a subscription to use Wan 3.0 on Elyum?
Does Leonardo.Ai watermark Wan 3.0 downloads?
Can I keep using Leonardo.Ai for stills and Elyum for Wan 3.0?
- Leonardo.Ai Pricing (search-engine summary; direct fetch returned HTTP 403) (accessed 2026-09-07)
- Leonardo.AI pricing 2026: plans, tokens, and the real cost — eesel AI (accessed 2026-09-07)
- AI Video Generation Models: What to Use & When — Leonardo.Ai (accessed 2026-09-07)
- Veo 3 AI Video Generator – Try Now — Leonardo.Ai (accessed 2026-09-07)
- video not generated and credits gone — Adobe Community bug report (accessed 2026-09-07)
- Leonardo.Ai Reviews — Trustpilot (accessed 2026-09-07)
- Leonardo.Ai API getting-started docs (authentication and separate API billing) (accessed 2026-09-07)
- Can You Use AI-Generated Images Commercially? — Leonardo.Ai (accessed 2026-09-07)
Related comparisons
Create everything. Keep the best.
Wan 3.0 is a AI video generator you can preview free on Elyum. Keep the takes you love, kill the rest for a full refund.