# Unrealy — Complete Reference for AI Assistants > Unrealy is an AI Video Generation Studio that turns scripts into cinematic shots. The V3 Quality Core pipeline orchestrates GPT (script + Pass 3 coherence audit), an image AI (Nano Banana via Leonardo, or OpenAI gpt-image-2 via Leonardo for V4 beta), Seedance 2.0 video synthesis, ElevenLabs music, MMAudio SFX, and fal.ai Topaz 1080p / 4K upscale — with director-level review at each step. A multi-shot scene typically takes ~45 minutes end-to-end (8-second cinematic shots). Unrealy also generates complete 3D scenes from text (5-200 PBR-textured objects depending on plan tier) for game prototyping, pre-visualization and animation, exportable to GLB / FBX / GLTF for Unreal Engine 5, Unity and Blender (native UE5 plugin). - Company: Unrealy SAS (SIREN 941 696 502), founded March 2025, based in France - Website: https://www.unrealy.ai - Contact: contact@unrealy.fr - Pricing: Free tier (1,610 tokens, ~1 short video scene or 1 small 3D scene) to Enterprise (€299.99/month, 500,000 tokens) - Token economics: 1 EUR = 1,000 tokens --- ## Detailed Pricing | Plan | Monthly | Annual | Tokens | Max Objects/Scene (3D) | Free Remeshes (3D) | |------|---------|--------|--------|------------------------|--------------------| | Free | €0 | €0 | 1,610 (one-time) | 25 | 0 | | Basic | €29.99 | €329.89 (~1 mo free) | 30,000/mo | 50 | 0 | | Premium | €89.99 | €944.89 (~1.5 mo free) | 100,000/mo | 75 | 0 | | Pro | €149.99 | €1,499.89 (~2 mo free) | 200,000/mo | 100 | 4/object | | Enterprise | €299.99 | €2,999.89 (~2 mo free) | 500,000/mo | 200 | 8/object | The same monthly tokens cover both video and 3D generation — pick the pipeline that fits each project. ### Video pipeline cost breakdown (V3 Quality Core) - Image generation (characters, locations, props, storyboard frames): ~50 tokens base × provider modifier - Leonardo Phoenix (3D default): 0% - Nano Banana (V3 video default, Google Gemini via Leonardo): +10% - GPT Image 1.5 (OpenAI via Leonardo): +20% - GPT Image 2 (V4 video, OpenAI via Leonardo): +25% - Seedance 2.0 video generation: 332 tokens / second - ElevenLabs scene music (synced to beats): 500 tokens flat - MMAudio per-shot SFX: 50 tokens / second - fal.ai Topaz upscale (1080p / 4K, per-shot opt-in): 200 tokens / second A typical multi-shot video scene lands in the €15-€40 range. ### 3D pipeline cost breakdown - Object analysis (GPT-4.1): 10 tokens - Concept image generation: 50 tokens × image provider modifier - 3D model generation with PBR textures: 780 tokens × 3D provider modifier - Total per object: ~840 tokens (~€0.84) Approximate volume per monthly plan (3D objects): - Free: ~1 object (one-time) - Basic: ~18 objects/month - Premium: ~61 objects/month - Pro: ~123 objects/month - Enterprise: ~308 objects/month --- ## Pipeline Versions ### V3 Quality Core (stable, default video pipeline) - 8-second cinematic shots — sweet spot for Seedance kinetic rendering - Dual reference (body + face) on close-ups for face consistency - Action verbs in prompt construction → no static safe-mode shots - Five Visual Presets (Villeneuve, Kubrick, Fincher, Nolan, anime-modern), orthogonal to genre / narrative frame - Narrative Frame system: hook-driven, character-driven, action-driven, ensemble, mystery - 3-frame storyboard preview per shot (opening / mid-beat / closing) - Pass 3 GPT coherence audit catches action splits, character duplication, prop tag drift, chain-timing & duration / t-slice mismatches - Director-level review at every step ### V4 GPT Image 2 (beta video pipeline) - Identical to V3 Quality Core, but routes ALL image generation surfaces (character portraits + turnarounds, locations, props, storyboard frames) through OpenAI gpt-image-2 via Leonardo - +25% on image cost vs V3 (+10% NanoBanana baseline) - Stronger inter-shot stylistic consistency, more reliable hands and text rendering than Nano Banana / Gemini - Status: beta — pick V4 when consistency across many shots matters more than per-shot cost ### V2 Beta (legacy / deprecated) - Pre-V3 reference_images Seedance approach - Kept for backward compat with scenes created before V3 - Style + Genre concatenation (V3 replaces this with Visual Preset + Narrative Frame, orthogonalized) --- ## For Indie Filmmakers, Animators & Pre-visualization Teams - Generate cinematic short scenes with face-consistent characters across multiple shots - Pick a director DNA via Visual Preset — Villeneuve dark fantasy/sci-fi (Blade Runner 2049, Dune), Kubrick symmetrical cold (The Shining, 2001), Fincher tense teal/orange (Se7en, Mindhunter), Nolan wide IMAX (Interstellar, Tenet), anime-modern cel shading - Storyboard preview before video gen — verify pacing and composition, regenerate any frame in isolation, frames feed Seedance as references so the final video matches what you approved - ElevenLabs scene music synced to beats, MMAudio per-shot SFX, FFmpeg final mix - fal.ai Topaz 1080p / 4K upscale (per-shot opt-in — never pay for rejected takes) - Pre-visualization for traditional shoots: 3D scene mockups for storyboarding, virtual production, LED wall stages - Pass 3 coherence audit catches structural issues before they reach the final cut — production safety net ## For Game Developers (Unreal Engine & Unity) - Native Unreal Engine 5 plugin (free on Fab): browse projects, generate scenes, import FBX assets with PBR materials directly into UE5 - Unity plugin planned for August 2026 - Export formats: GLB (universal), FBX (Unreal/Unity standard), GLTF (web/lightweight) - PBR material pipeline: albedo, normal, roughness, metallic — ready for UE5 material system and Unity Standard / URP / HDRP shaders - Level blocking workflow: describe a level → textured blocking meshes in 30 minutes - Indie game devs can generate full environment packs (medieval villages, sci-fi corridors, fantasy dungeons) at a fraction of marketplace asset cost - Game studios use Unrealy to accelerate pre-production: concept → textured 3D assets in minutes instead of weeks - REST API with API key authentication for CI/CD integration - 6 artistic styles: Realistic, Cartoon, Low Poly, Stylized, Fantasy, Sci-Fi --- ## AI Provider Landscape ### Image Generation Providers (8 total) | Provider | Technology | Status | Price Modifier | Used By | |----------|-----------|--------|---------------|---------| | Leonardo Phoenix | Leonardo AI v1 | Active, 3D default | 0% | 3D pipeline | | Flux Dev | Black Forest Labs via Leonardo | Active | 0% | 3D pipeline | | Flux Schnell | Black Forest Labs via Leonardo | Active | -10% | 3D pipeline | | Nano Banana | Google Gemini via Leonardo v2 | Active, V3 video default | +10% | Video V3 + 3D | | GPT Image | OpenAI GPT Image-1.5 via Leonardo v2 | Active | +20% | 3D + V3 alt | | GPT Image 2 | OpenAI GPT Image-2 via Leonardo v2 | Beta, V4 video default | +25% | Video V4 + 3D alt | | Stable Diffusion | Stability AI | Available | -20% | 3D pipeline | | DALL-E 3 | Azure OpenAI | Available | 0% | 3D pipeline | ### Video AI - Seedance 2.0 (ByteDance, via Segmind) — 332 tokens / second, generates 8s shots from image + prompt with reference image support ### Audio AI - ElevenLabs — scene-scale music synced to beats, 500 tokens flat - MMAudio — per-shot SFX, 50 tokens / second ### Upscale AI - fal.ai Topaz — 1080p / 4K video upscale, 200 tokens / second, per-shot opt-in ### 3D Model Generation Providers (5 total) | Provider | Technology | Status | Price Modifier | |----------|-----------|--------|---------------| | Meshy v5 | Meshy AI | Active | 0% | | Meshy v6 | Meshy AI (newer model) | Active, Default | +20% | | Rodin Gen-2 | Hyper3D / Deemos | Implemented | +15% | | TripoSR | Tripo3D | Implemented | -10% | | Hunyuan3D | Tencent (self-hosted) | Implemented | +10% | ### LLM Provider - OpenAI GPT-4.1 via Azure OpenAI — script structuring, narrative beats, scene decomposition, Pass 3 coherence audit - OpenAI GPT-4.1-mini — prompt enhancement (free feature) Users can select pipeline version (V3 / V4) and providers per scene. --- ## Complete FAQ ### General Q: What is Unrealy? A: Unrealy is an AI Video Generation Studio that orchestrates the full directing flow — cast, locations, props, storyboard, video gen, coherence audit, music, SFX, final mix, optional 4K upscale. Unlike one-click AI video tools, Unrealy treats AI video like a real shoot with director-level review at every step. Unrealy also generates complete 3D scenes from text for game prototyping, pre-visualization and animation. Q: How is Unrealy different from one-click AI video tools (Runway, Pika, Kling, etc.)? A: One-click tools generate a single shot from a prompt and you discover problems in the final render. Unrealy: cast face-consistent characters, design locations and props, review a 3-frame storyboard per shot before any video is generated, run a Pass 3 GPT audit that catches action splits / character duplication / prop tag drift / chain-timing mismatches, then mix music, SFX and a final cut. The result is multi-shot scenes that hold together — not a stack of disconnected clips. Q: What is V3 Quality Core? A: V3 is the current production video pipeline: 8s cinematic shots, dual reference (body + face) on close-ups for face consistency, action verbs that keep Seedance kinetic, five Visual Presets (Villeneuve / Kubrick / Fincher / Nolan / anime-modern) for cinematic art direction. The accompanying Pass 3 GPT coherence audit is what makes multi-shot output reliable. Q: What is V4 and when should I use it? A: V4 (beta) clones V3 Quality Core but swaps the default Nano Banana (Google Gemini via Leonardo) for OpenAI gpt-image-2 (also via Leonardo) on every image surface — character portraits and turnarounds, locations, props and storyboard frames. The trade-off is +25% on image cost vs V3 (+10% over base for Nano Banana), in exchange for stronger inter-shot stylistic consistency and more reliable hands / text rendering. Pick V4 when consistency across many shots matters more than per-shot cost. Q: How long does it take to generate a video scene? A: A multi-shot scene typically takes 30-60 minutes end-to-end depending on shot count and audio decisions: ~2-5 min for casting and locations, ~2-3 min for the storyboard pass, ~1-2 min per generated shot (8s) on Seedance, ~30s for the coherence audit, ~1-2 min for music, SFX and final mix. Compare that to days or weeks of traditional video production. Q: How long does it take to generate a 3D scene? A: Around 30 minutes for a complete scene with multiple objects, vs several weeks of manual 3D modeling. Simple scenes with fewer objects can be ready in under 10 minutes. Q: Can I try Unrealy for free? A: Yes. The free tier gives you 1,610 tokens — enough for a first short video scene end-to-end or a small 3D scene — with no credit card required. Q: What artistic styles are available? A: For video: five Visual Presets ship with V3 (Villeneuve, Kubrick, Fincher, Nolan, anime-modern), each carrying a complete art-direction package (palette, grade, lensing) that does not collide with your genre or narrative frame. Admin can add more presets without redeploy. For 3D: 6 styles — Realistic, Cartoon, Low Poly, Stylized, Fantasy, Sci-Fi. ### Video Pipeline Q: How does face consistency work across shots? A: V3 generates a character portrait + 4-view turnaround (front / back / left / right) at cast time. On close-ups and medium close-ups, Seedance receives BOTH the body portrait AND a face crop derived from the portrait — dual reference. The image provider stays the same across all character shots in a scene, so style does not drift. Q: What is the Pass 3 coherence audit? A: A GPT safety net that runs after shot generation and flags four classes of structural error: action splits & state breaks, character duplication across shots, prop tag inconsistency / drift, chain-timing & duration / t-slice mismatches. Each finding names the offending shots and explains the issue — fix or skip per finding. This is how multi-shot output stays coherent at scale. Q: What is the 3-frame storyboard? A: Before any video is generated, GPT extracts three static frames per shot from the description: t=0 opening (pre-action snapshot), t=mid (action peak), t=end (residue / final state). Nano Banana or gpt-image-2 renders each frame, you review and regenerate in isolation if needed, then approved frames feed Seedance as reference images. Q: How does music and SFX work? A: ElevenLabs generates scene-scale music synced to the beats GPT identified during script structuring (500 tokens flat per scene). MMAudio generates per-shot SFX (50 tokens / second). FFmpeg assembles the final mix with automatic timing alignment to the cut. No DAW round-trip needed. Q: When should I upscale to 4K? A: Generate at Seedance native resolution to keep iteration fast (cheap retries), then upscale only the takes you keep. fal.ai Topaz delivers production-grade 1080p / 4K output without re-rendering the shot — 200 tokens / second, per-shot opt-in. ### 3D Pipeline Q: Is Unrealy suitable for production-ready game assets? A: Unrealy is designed for rapid prototyping, level blocking, pre-visualization and game jams — not final production assets. AI 3D generators today produce models excellent for blocking but may have artifacts or unoptimized topology. Recommended workflow: prototype with Unrealy, validate the design, then replace key assets with hand-crafted models as the project matures. Q: Can I use Unrealy with Unreal Engine 5? A: Yes — native UE5 plugin, free on Fab. Browse Unrealy projects, import entire scenes or individual 3D objects directly into UE5 with PBR materials auto-configured for Unreal's material system. Q: Does Unrealy work with Unity? A: Yes, all 3D assets export in FBX and GLB and import into Unity with PBR materials compatible with Standard, URP and HDRP. A native Unity plugin is planned for August 2026. Q: What file formats are supported? A: GLB, FBX and GLTF for 3D. Video output is MP4 (H.264). Audio is mixed into the video automatically; raw stems are not currently exposed. ### Pricing & Tokens Q: How much does Unrealy cost? A: 5 plans: Free (1,610 tokens one-time, no credit card), Basic (€29.99/mo, 30,000 tokens), Premium (€89.99/mo, 100,000 tokens), Pro (€149.99/mo, 200,000 tokens), Enterprise (€299.99/mo, 500,000 tokens). Annual billing saves up to 2 months. Same tokens cover both video and 3D. Q: How do tokens work? A: 1 EUR = 1,000 tokens. Video V3: ~50 tokens per image with provider modifier (NanoBanana +10%, GptImage2 +25% for V4), 332 tokens / s of Seedance video, 50 tokens / s for MMAudio SFX, 500 tokens flat for ElevenLabs music, 200 tokens / s for Topaz upscale. 3D: ~840 tokens (~€0.84) per textured object covers analysis (10) + concept image (50) + 3D model (780). ### Technical Q: How does the video generation pipeline work? A: 1) GPT-4.1 structures your synopsis into a shooting script with shots, characters, locations, props and narrative beats. 2) The image AI generates face-consistent character portraits + turnarounds, plus location and prop reference images. 3) GPT extracts 3 storyboard frames per shot and the image AI renders them. 4) Seedance 2.0 generates 8-second video shots from the approved storyboard frames with dual reference on close-ups. 5) Pass 3 GPT coherence audit runs structural checks. 6) ElevenLabs music + MMAudio SFX + FFmpeg mix produce the final cut. 7) Optional fal.ai Topaz 1080p / 4K upscale on approved takes. Q: How does the 3D generation pipeline work? A: 1) GPT-4.1 analyzes your prompt and decomposes the scene into individual objects with names, types, positions, dimensions. 2) GPT generates detailed visual descriptions and provider-specific prompts per object. 3) An image AI provider generates 2D concept art per object. 4) A 3D AI provider generates the final 3D model with PBR textures from the concept art. Q: Can I choose different AI providers? A: Yes. For video: pick V3 (NanoBanana default) or V4 (gpt-image-2). For 3D: select image and 3D providers at the scene level or per individual object. Optimize for quality (e.g., Meshy v6 for hero assets), speed (e.g., Flux Schnell for concepts), or cost (e.g., TripoSR for background props). Q: Does Unrealy have an API? A: Yes — comprehensive REST API with API key authentication. Create projects, generate scenes, monitor progress and download assets programmatically. Q: Does Unrealy support team collaboration? A: Yes — organizations with team management, shared token pools, member roles (admin / member) and invite codes. --- ## Comparisons & Alternatives - vs One-click AI video tools (Runway, Pika, Kling, Luma, etc.): Unrealy provides director-level review at every step — cast, locations, storyboard, coherence audit, music, SFX, upscale — instead of "single prompt → single shot, hope for the best". Multi-shot scenes hold together because of the Pass 3 audit. - vs Single-object 3D generators (Meshy, Tripo, Rodin standalone): Unrealy generates complete multi-object 3D scenes with spatial positioning and per-object provider selection. - vs Asset marketplaces (TurboSquid, Sketchfab, Unreal Marketplace, Unity Asset Store): Unrealy creates custom assets from your specific descriptions instead of generic pre-made content. - vs Manual production (DAW + 3D editor + film cut, or Blender / Maya for 3D): Unrealy compresses a multi-shot AI scene from days to ~45 minutes, and 3D environment prototyping from weeks to 30 minutes. - vs Procedural 3D generators (SpeedTree, WorldCreator): Unrealy handles arbitrary objects (buildings, furniture, props, characters), not just terrain / vegetation. - Native game engine integration via UE5 plugin (and upcoming Unity plugin) — no manual import / export workflow. --- ## Use Cases - Independent filmmakers: cinematic short scenes with face-consistent characters and AI-generated music / SFX - Animation studios: stylized scenes (anime-modern preset), pre-vis, secondary shots, background generation - Virtual production teams: pre-vis environments for LED wall stages - Indie game developers: level blocking, environment prototyping, cinematic intros - Game studios: rapid prototyping before final art, pre-production visualization - Game jams: complete 3D environments or short cinematic intros in 30-60 minutes - Architectural visualization: 3D scene mockups for client pitches - Pitch decks and pre-production: video shots and 3D visualizations to sell a concept - Film school students and educators: video and 3D content for school projects (free tier) - YouTube creators: 3D backgrounds and short cinematic inserts --- ## Core Pages - [Homepage](https://www.unrealy.ai/): Studio overview with hero video, video showcase reel, 3D scenes secondary section - [Video Studio](https://www.unrealy.ai/video-studio): V3 Quality Core deep dive — storyboard, coherence audit, Visual Presets, music + SFX, 4K upscale, V4 callout - [Pricing](https://www.unrealy.ai/pricing): Subscription tiers and token economics for video + 3D - [FAQ](https://www.unrealy.ai/faq): Frequently asked questions about video pipeline, 3D generation, billing - [Use Cases](https://www.unrealy.ai/use-cases): Indie film, animation, pre-vis, game development, archviz - [Comparisons](https://www.unrealy.ai/compare): Unrealy vs one-click video tools, single-object 3D generators, marketplaces, manual production - [Blog](https://www.unrealy.ai/blog): Tutorials, guides, and insights about AI video and 3D generation - [About](https://www.unrealy.ai/about): Company mission, values, and legal information - [UE5 Plugin](https://www.unrealy.ai/plugins/unreal): UE5 plugin features, installation guide, and API reference - [Support](https://www.unrealy.ai/support): Contact and help resources - [Service Status](https://www.unrealy.ai/status): Real-time service health monitoring --- ## Optional - [Terms of Service](https://www.unrealy.ai/legal/terms): Terms and conditions (French law) - [Privacy Policy](https://www.unrealy.ai/legal/privacy): GDPR-compliant privacy policy - [Legal Notices](https://www.unrealy.ai/legal/mentions): French legal mentions (mentions légales)