> For the complete documentation index, see [llms.txt](https://wiki.secrets.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://wiki.secrets.ai/create/video-generation.md). # Video Generation ### How Video Generation Works {% hint style="info" %} Video generation requires a **PREMIUM**, or **ULTIMATE** subscription. Pricing depends on your **content mode**, **resolution**, **duration**, and **audio settings**. The exact cost is always shown on the Generate button before you generate. {% endhint %} Video generation takes a generated image and animates it based on your prompt, creating a short video clip with motion, effects, and life. {% hint style="info" %} **New in the Generation Studio:** video generation now happens **right on the Generate page** — no separate window. Use the **Image / Video toggle** at the top of the generation form to switch between generating images and videos. {% endhint %} #### The Process 1. **Start with an Image**: Switch to **Video** on the Generate page, or tap the video icon on any generated image — from the Generate page gallery or your **Collection page** 2. **Choose Content Mode**: Advanced (Quality or Fast), Standard, or Adult 3. **Set Your Frames or References** (Advanced only): Pick a first frame, an optional end frame, or blend multiple reference photos and clips 4. **Enter Video Prompt**: Describe the motion and effects you want 5. **Generate Video**: AI creates a video based on your image and prompt 6. **View Result**: Watch your animated video in the gallery
#### Content Modes & Settings Video generation supports three content modes: * **Advanced**: The primary video generation model, with reference media, frame control, and audio always on. Comes in two variants: * **Quality**: The highest fidelity results, with resolutions up to **4K** * **Fast**: Quicker generations, with resolutions up to 720p * **Standard**: The legacy SFW model with optional audio * **Adult**: NSFW video generation (720p, 8 seconds, no audio) | Mode | Resolutions | Duration | Audio | | ---------------------- | --------------------- | ------------ | ------------- | | **Advanced (Quality)** | 480p, 720p, 1080p, 4K | 4–15 seconds | Always on | | **Advanced (Fast)** | 480p, 720p | 4–15 seconds | Always on | | **Standard** | 480p, 720p, 1080p | 4–12 seconds | Optional | | **Adult** | 720p | 8 seconds | Not available | {% hint style="info" %} **4K video is exclusive to Advanced (Quality).** 720p is the recommended resolution for the best balance of quality, speed, and cost. {% endhint %} ### Reference Media & Frame Control {% hint style="info" %} Reference media and frame controls are available on **Advanced** video generation (both Quality and Fast). Adding reference media costs **no extra moments** — the price depends only on your resolution, duration, and mode. {% endhint %} Advanced mode gives you two ways to control your video, switchable via the **Frames / References** tabs: #### Frames Control exactly where your video starts and ends: * **First frame**: The image your video starts from. Set only a first frame and the model animates it based on your prompt. * **End frame (optional)**: Add an end frame and the video will transition from your first frame to your last frame — great for reveals, outfit changes, and camera moves with a clear destination. #### References Blend multiple pieces of media to steer the generation: * Up to **5 reference photos** (your source image counts as the first one) * Up to **3 reference clips**, with a combined length of up to **15 seconds** The model uses your references to guide appearance, style, and motion rather than starting from a single fixed frame. #### Picking Reference Media The reference picker pulls from your **entire gallery** — every image and video you've generated. Filter by companion, tags, favorites, or most recent to find exactly what you need. ### Video Prompting Guide {% hint style="info" %} Advanced mode prompts can be up to **5,000 characters** long. The model has excellent prompt adherence — the more detail you provide about the motion you want, the better your results will be. {% endhint %} The Advanced video model is a massive leap in quality. It handles complex motion, realistic physics, multi-subject interactions, and synchronized audio generation all in a single pass. To get the most out of it, follow these prompting principles. #### Core Prompt Structure Since the image already provides the visual context, build your prompt around motion and change: 1. **Action & motion** — What moves and how (the most important element) 2. **Camera movement** — How the "camera" moves (one movement per shot) 3. **Pacing** — How fast or slow things happen 4. **Audio cues** — Dialogue, sounds, or atmosphere you want to hear 5. **Constraints** — What to avoid (no distortion, no jitter, etc.) {% hint style="info" %} **Aim for 50–200 words.** This is the sweet spot for quality. You don't need to re-describe what's already in the image — focus on the animation. {% endhint %} #### Key Principles **1. Focus on motion, not the scene** The image is your scene. Don't re-describe the setting, outfit, or appearance — the model already sees all of that. Instead, describe what *changes*. * ✅ "She slowly turns her head and smiles, hair swaying gently in a breeze" * ✅ "Leans forward slightly, eyes narrowing with a playful expression" * ❌ "A woman standing on a beach in a red dress at sunset" — the image already shows this **2. Be specific about intensity and pacing** Use degree words to control how movements feel. The model is very responsive to pacing cues. * ✅ "Slowly brushes hair behind her ear" * ✅ "Gentle breeze lifts her hair, subtle movement" * ✅ "Quickly glances over her shoulder" * ❌ "Moves" — too vague, unpredictable result **3. One camera movement per shot** The model handles camera movement well, but only when you give it one clear instruction. Combining multiple movements causes jitter. * ✅ "Slow dolly push-in" * ✅ "Smooth pan to the right" * ✅ "Camera holds still" — perfectly valid, keeps the focus on subject motion * ❌ "Dolly in while panning left and tilting up" — too many movements at once Use pacing words like "slow," "smooth," or "gentle" rather than technical parameters. **4. Separate camera movement from subject movement** This is the most common prompting mistake. Be explicit about what moves and what stays still. * ✅ "She slowly turns her head to the right. Camera holds fixed framing." * ✅ "Camera slowly pushes in. She holds her pose, only her hair moves in the wind." * ❌ "Everything moving at once" — confuses the model **5. Use sequential actions for complex scenes** List actions in the order you want them to happen. The model follows temporal sequences well: * "Looks down at her phone with a surprised expression, then looks up at the camera with excitement, gasps softly" * "Takes a sip of coffee, sets the cup down, then glances out the window with a thoughtful expression" **6. Add quality constraints at the end** Append a short constraint line to any prompt for consistently better output: * "No distortion. No jitter. Face stable, no deformation." This works across all content types and noticeably reduces common artifacts. #### Example Prompts **Simple & effective:** > She brushes hair behind her ear with a soft smile, gentle breeze, smooth and natural motion. No distortion. **With camera movement:** > Slow dolly push-in. She looks up toward the camera with a warm, inviting expression, lips parting slightly. Hair shifts gently. No jitter, face stable. **Complex sequential action:** > She glances down at her phone and her eyes widen with surprise, then she looks up at camera with excitement and laughs softly. Camera holds still. No distortion, no deformation.
#### Prompting With Audio Audio is generated alongside the video in Advanced mode — not as a separate step. The model reads the source image and your prompt to produce matching sound automatically. **Three audio layers are generated simultaneously:** * **Lip-synced speech** — Include short dialogue in your prompt with emotional context for best results (e.g., "She softly whispers 'Hey, I missed you'"). Keep dialogue to **5–10 words per line** for the cleanest lip sync.
* **Sound effects** — Tied to the actions you describe. Footsteps, cloth movement, glass clinking, a sigh — the model generates these based on what's happening in the scene.
* **Ambient sounds & music** — Background atmosphere inferred from the source image and prompt (city noise, ocean waves, quiet room tone, soft background music). {% hint style="info" %} **Audio tip:** Label emotions before dialogue for much better results. "She whispers softly" produces noticeably better lip sync and tone than just writing the dialogue alone. {% endhint %} #### Things to Avoid | Don't do this | Why | Do this instead | | ------------------------------------- | ----------------------------------------------------------------------------- | ---------------------------------------- | | Re-describing the image | The model already sees the image — redundant prompting wastes space | Focus on motion and change | | Multiple camera movements in one shot | Causes jitter and incoherent motion | One clear camera movement per shot | | Contradicting the source image | Creates visual confusion (e.g., describing a beach when the image is indoors) | Stay consistent with what's in the image | | Overloading a short clip | 4–15 seconds can't contain an entire story | One clear moment per generation | | Vague motion ("she moves") | Unpredictable results | Be specific about what moves and how | | Long dialogue lines | Lip sync degrades with long speech | Keep dialogue to 5–10 words | {% hint style="info" %} **Pro tip:** For clips longer than \~8 seconds, consider whether two shorter generations might give you better results. Quality stays highest in the 4–8 second range. {% endhint %}
### Requirements & Limitations #### Subscription Requirements * **FREE Tier**: ❌ Not available * **PREMIUM Tier**: ✅ Available * **ULTIMATE Tier**: ✅ Available ### Understanding Video Results #### What to Expect The Advanced model handles complex motion, realistic physics, and multi-subject interactions far better than previous models. That said, AI video generation still has some limitations: * **Hands & fingers** — Occasional extra or fused digits can appear in complex hand movements * **Fine patterns** — Tight weaves, tiny checks, and detailed textures can shimmer or flicker. Simpler patterns work better. * **Variation** — Results vary between generations, even with the same prompt * **Iteration** — You may need to regenerate or tweak your prompt for the best result ### Frequently Asked Questions **Q: How much do videos cost?** * **Advanced**: Cost scales with resolution and duration — the exact price is shown on the Generate button before you generate. Audio is always on and included. * **Standard**: Cost scales with resolution, duration, and whether audio is enabled. * **Adult**: 600 moments (720p, 8 seconds). * Adding reference photos or clips costs nothing extra. **Q: Do I need a subscription?** * Yes, video generation requires a PREMIUM or ULTIMATE subscription. **Q: Can I use any image?** * Yes, any generated image can be used as source for video. **Q: What resolutions can I generate?** * **Advanced (Quality)**: 480p, 720p, 1080p, or 4K. * **Advanced (Fast)**: 480p or 720p. * **Standard**: 480p, 720p, or 1080p. * **Adult**: 720p. **Q: How do I generate in 4K?** * Choose **Advanced** mode and the **Quality** variant — 4K is exclusive to Advanced (Quality). **Q: What's the difference between Quality and Fast?** * **Quality** produces the highest fidelity results and supports up to 4K. **Fast** generates quicker and tops out at 720p. Both support the full 4–15 second range, reference media, frame control, and audio. **Q: Can I use reference photos and clips?** * Yes! In Advanced mode, switch to the **References** tab to blend up to **5 reference photos** and **3 reference clips** (15 seconds of clip time combined) from your gallery. It costs nothing extra. **Q: Can I control the first and last frame?** * Yes! In Advanced mode, use the **Frames** tab to set a first frame, and optionally an end frame — the video will transition from one to the other. **Q: Can I add audio?** * In **Advanced** mode, audio is always on. In **Standard** mode, audio is optional. * Audio supports speech lip sync, ambient sounds, action sounds, and background music. **Q: Can I download videos?** * Yes, you can download generated videos from the videos tab. **Q: Can I generate multiple videos from same image?** * Yes! Generate as many videos as you want from the same image with different prompts. **Q: What motion can I create?** * The Advanced model excels at complex, realistic motion — walking, turning, hair blowing in wind, clothing movement, facial micro-expressions, multi-subject interactions, and much more. It handles physics naturally, so don't be afraid to get creative. **Q: How do I get the best results?** * Start your prompt with the shot type, always describe the lighting, use one camera movement per shot, and add "No distortion. No jitter." at the end. See the [Video Prompting Guide](#video-prompting-guide) for detailed tips. *** Want to learn more? Check out [Image Generation](/create/image-generation.md) or explore [Custom Characters](/create/custom-characters.md)!