How to use izgens
Everything you need: engines, references, prompts, photo, tools. 4 minute read.
E Video engines
Pick the engine from the dropdown at the top of the video studio. Different engines mean different models, price points, and strengths — the rest of the form (refs, prompt, options) adapts to each.
🎬 Seedance 2.0 Standard DEFAULT
Main video engine. Best identity preservation and cinematic look. Use for products, portraits, animation, brand shots. Strict on real-person adult content — switch to NSFW variant when needed.
🔞 Seedance 2.0 NSFW
Same core model with looser content rules for adult work. Auto-adds +15% to base cost.
⚡ Seedance 2 Mini CHEAP
Faster, cheaper Seedance. Best for previews, ideation, quick tests before committing to a full Standard run. Available in NSFW too.
🌊 Wan Video 2.7
Alternate video model. Two modes: I2V (image → video, one image only) and R2V (reference → video, up to 5 images + a video ref). Simpler options, no lip-sync or generate-audio toggles.
🎭 Kling Motion
Motion transfer engine. Upload one character image + a motion video: the character adopts the video's movement. Two orientations — Video (up to 30s, follows motion clip's facing) and Image (up to 10s, keeps portrait's facing). Aspect ratio is auto-derived from the video.
🎬 Kling 3.0 Video
Prompt + image → video. First image = start frame, optional second = end frame. Supports up to 4K output. Ambient sound toggle available. Best when you want higher resolution than Standard can deliver.
Rule of thumb: start with Standard for anything final. Use Mini for A/B testing prompts. Switch to NSFW only when Standard rejects your content. Try Wan for alternate look, Kling Motion when you have a driving clip, Kling 3.0 for 4K finals.
R Reference slots
Each engine adapts the ref slots to what it accepts. The bar under "References" always tells you how many images/videos are allowed for the current engine.
1
Image 1 — Primary. Required. Behaves two ways:
• Alone or with more images → treated as the start frame or main subject. The model animates outward from it. Match the aspect ratio you'll pick.
• With a video reference → treated as an identity reference. The model takes the person/object from here and drops them into the scene from the video. Doesn't have to be the first frame — a clean portrait works.
2
Image 2–5 — Extra identity or scene refs. Optional. Add a front-facing portrait as Image 2 to lock the face when Image 1 is a wide shot. Seedance uses up to 5, Wan R2V up to 5, Wan I2V just 1, Kling Motion just 1.
V
Video reference. Optional (Seedance / Wan R2V / motion) or required (izgens motion, Kling Motion). 2–15 seconds, MP4. The model lifts motion, timing, and scene composition from it. Great for "take person from @img1, do what @vid does." Small cost bump when used.
Quality in, quality out. Refs under 512px, heavily compressed screenshots, and photos with hard shadows or busy backgrounds all reduce identity fidelity. Bright, clean, front-facing portraits work best.
P Prompting
Prompts describe what happens, not who's in the shot — the references already show that. Keep it tight, specific, and action-first.
Core rules
→
Describe action, not appearance. "Slowly turns head left, blinks, half-smile" beats "beautiful young woman with brown hair." The image already shows the person.
→
Keep it under 60 words. Long prompts dilute attention. One scene, one camera move, one main action per prompt.
→
Front-load the important thing. The model weighs early tokens more heavily. Lead with the action, add style/lighting after.
→
Camera language works. "Slow dolly in", "handheld tracking shot", "static wide", "orbit right around subject" — these are understood.
@ mentions
@
Type @ anywhere in the prompt to open a picker. Insert @img1, @img2, @vid to point the model at a specific reference. Use this when you have multiple refs and want to be explicit about which one drives what.
@
Example: "@img1 walks into the scene from @vid, matches the lighting." — this tells the model "put the person from image 1 into the scene from the video, and blend them into the video's lighting."
Example prompts
@img1 slowly turns head to the right, gentle smile, blinks once. Static camera, soft window light from left. 3-second calm beat.
Take the person from @img1 and place them into the scene from @vid, mirroring the exact pose and camera motion. Lock facial identity to @img2. Photorealistic, no style shift.
Wide cinematic shot: @img1 walks toward camera on an empty beach at sunset, hair moved by wind. Slow dolly-in. Warm golden-hour tones, shallow depth of field.
Presets
★
Built-in presets — click "Presets" in the Prompt panel. "Character animation (full)" and "Talking head with lip-sync" auto-fill battle-tested templates. Edit from there.
★
Save your own. Open
Settings → Prompt Templates. Anything you save shows up under "My presets" in the Prompt panel dropdown. Reuse across sessions.
Content filter tripped? Rewrite explicit words as descriptions of action or setting. "Naked" → "in swimwear on a private beach". "Sexy" → "confident, direct eye contact". If it still fails, switch to the NSFW variant of your engine. Failed generations are auto-refunded.
O Options that matter
⏱
Duration. 5s for quick tests (cheapest), 10s standard, 15s for full shots. Pricing scales roughly linearly.
📐
Aspect ratio. 9:16 for TikTok/Reels/Shorts, 16:9 for YouTube/hero, 1:1 for feed posts, 3:4 / 4:3 for older formats. Match this to your reference's aspect for best framing.
🖥
Resolution. 480p for ideation, 720p is the sweet spot for social, 1080p for finals. Each step up adds roughly 60–80% cost.
🎙
Generate audio (Seedance only). AI-built ambient/effects soundtrack matching the scene. Off by default for talking heads.
💋
Lip-sync (Seedance only). Pair with a voice clip (English works best) to sync mouth movement. Best paired with the "Talking head" preset. Auto-turns Generate audio on when enabled.
🧑
Real person mode (Seedance Standard only). Turn ON when your reference is a real photo (not anime / 3D). Improves identity preservation, applies our real-person bypass.
🧼
Clean metadata (+5%). Strips invisible AI watermarks and EXIF/identifiers before download. Turn on when posting to TikTok / Instagram / Reels.
Engine-specific toggles. The options row auto-adapts: Wan Video hides lip-sync, generate audio, and real-person; Kling Motion shows only Character Orientation. Anything hidden isn't accepted by that engine.
P Photo studio
Generate or remix images. Up to 10 reference images per prompt. Pick the model from the dropdown — each has different strengths.
🖼 Seedream 4.5 DEFAULT
Fast, cheap, strong at portraits, products, and edits. Best default for most work. NSFW variant available.
💎 Seedream 5.0 Pro
Higher-fidelity, better prompt adherence for complex scenes. Costs more per image. Use when 4.5 misses details.
🍌 Nano Banana 2 / Pro
Excellent at photorealism and prompt understanding, especially for scenes and object relations. Pro = higher quality tier.
🌊 Wan 2.7 Image
Good stylistic range, alternative aesthetic compared to Seedream. NSFW-tolerant.
Workflow
1
Add references — drag-drop or click "+". Each gets a tag @img1, @img2 etc. Up to 10.
2
Reorder by drag. The order is the order the model reads them.
3
Write a prompt, use @ to reference specific images. Example: "Combine outfit from @img2 with pose from @img1, background from @img3."
4
Pick size, aspect, model. 2K for social, 3K/4K for print. NSFW variants available on Seedream 4.5 / 5.0 / Wan.
5
Generate. Click a tile to open full-size; recreate or download from there. You can send any photo straight to Photo Upscale via the Tools button.
Face consistency is best on Seedream 5.0 Pro and Nano Banana Pro. If a character keeps drifting, upload 2–3 different angles of them as references and mention them explicitly in the prompt.
$ Credits
Everything runs on credits. Top up via card or crypto — bigger packs come with bonus credits.
✓
Cost preview. Every Generate button shows the exact credit cost before you click. No surprise charges. Duration, resolution, and options each shift the number in real time.
✓
Mini is cheaper. Use Seedance Mini for A/B testing prompts, then re-run the winner on Standard.
✓
Failed generations are auto-refunded. Content-filter rejects, timeouts, upstream errors — all return credits automatically. Check
Account if anything looks off.
✓
Promo codes. Referral partners have codes that grant bonus credits on your first top-up. Enter one on the Checkout page if you have it.
izgens.
← Back to studio