> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meigen.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Image & Video Generator
> Generate images & videos on MeiGen with GPT Image 2, Nano Banana 2, Seedream 5.0 Lite, Midjourney V8.1, Seedance 2.0, Veo 3.1 & more. Prompt editor with variables, color picker, and @ mentions.
MeiGen lets you generate images and videos using multiple AI models from a unified interface. On desktop, open the generation sidebar from the floating dock; on mobile, tap **Generate** in the bottom dock to open the generation sheet. Both surfaces offer the same models, reference images, `@` mentions, prompt enhancement, AI edit, and Storyboard — this guide applies to both unless a section says otherwise.
## Model Selector
At the top of the generation sidebar, the model selector dropdown lets you choose which AI model to use.
Each option shows:
* **Provider icon** — visual indicator of the model provider (Google, ByteDance, Midjourney, etc.)
* **Model name** — the display name of the model
* **Credit cost** — how many credits this model uses per generation
Selecting a different model automatically updates the available aspect ratios, resolution options, and advanced settings to match that model's capabilities.
Compare available models — features, pricing, strengths, and limitations
***
## Prompt Editor
### Writing Effective Prompts
* **Be specific** — describe composition, lighting, colors, textures, and perspective for the best results
* **Short prompts work too** — use the Enhance button to expand brief descriptions into detailed prompts automatically
* **Language** — most models accept English prompts. Midjourney V8.1 will auto-translate non-English prompts to English
### Variable Tags
Some prompts from the gallery contain blue `[placeholder]` tags — for example, `[style]`, `[subject]`, or `[color palette]`. These are **editable template variables**.
Click any blue tag to select it, then type your replacement text. This makes it easy to create variations of a prompt without rewriting the whole thing.
### @ Image Mentions
When you have multiple reference images uploaded, you can mention specific images inline in your prompt:
1. Type `@` in the prompt editor
2. A popup appears showing thumbnails of your uploaded reference images
3. Select an image to insert a mention tag (e.g., `@image1`, `@image2`)
4. Write instructions around the mentions: *"Use @image1 as the background scenery and apply the color palette from @image2"*
This gives you precise control over how each reference image influences the generation.
### Characters
Create a character once and reference it across generations, so the same figure stays consistent from image to image — and into video.
1. Click the **Characters** button above the prompt editor (available in both the image and video sidebars)
2. In **Create Character**, give it a name and add **1–4 images** — upload them, or pick from your recent creations
3. Back in the prompt, type `@` and select the character: *"@Mia standing in a neon-lit alley, cinematic lighting"*
You don't need to attach the character's images as reference images again — selecting it in the prompt is enough. Rename or delete characters under **My Characters**; deleting one breaks the prompts that reference it.
Best results: give the character one vertical full-body shot plus one neutral face close-up of the same character.
Only AI-generated or virtual characters are allowed. Images that resemble a real person are rejected by content review.
**Limits** — there is a cap on how many characters an account can keep, and on how many you can create per day. The dialog tells you when you reach it; delete unused characters under **My Characters** to free up room.
**Availability** — image models that accept reference images (all except Flux 2 Klein and Z Image Turbo), plus **Seedance 2.0** for video. Selecting a character on a text-only model returns an error.
If you submit a work that mentions a character to the public gallery, that character's main image becomes publicly visible alongside it — the publish dialog asks you to confirm first.
### # Color Picker
When you need to specify an exact color for an object in your prompt, pick one from the palette:
1. Type `#` in the prompt editor
2. A palette appears with common colors and curated hues grouped by shade
3. Click any swatch to insert a color tag (e.g., `#dc143c`)
4. Write instructions around the color: *"A t-shirt in `#dc143c` paired with `#f5f5dc` pants"*
Colors are submitted to the model as hex values — more precise than describing "deep red" or "beige" in words.
### Translate Button
Click the translate icon (top right of the prompt editor) to translate your prompt into English. This is useful when writing prompts in another language, as most models produce the best results with English input.
### Prompt Enhancement
Click the **Enhance Prompt** button to automatically enhance your prompt based on the selected model:
* **Polish mode** (most models) — preserves intent, adds visual details like composition, lighting, and materials
* **Expand mode** (Midjourney V8.1) — expands into Midjourney-optimized language; non-English prompts are auto-translated
```text Input theme={null}
a cat sitting on a windowsill
```
```text Polished theme={null}
A fluffy orange tabby cat sitting gracefully on a sunlit windowsill,
warm golden hour light streaming through sheer curtains, soft bokeh
of a garden visible outside, cozy interior with houseplants, photorealistic
style with shallow depth of field
```
Enhancement works best on brief prompts (under 30 words). Skip it for prompts that already have detailed visual descriptions — the AI may over-process them.
### Keyboard Shortcuts
| Shortcut | Action |
| -------------------------- | ------------------ |
| `Cmd+Enter` / `Ctrl+Enter` | Trigger generation |
***
## Reference Images
Upload reference images as visual context to guide the AI's output — for style matching, composition guidance, or subject reference.
* **Upload methods** — drag-and-drop files onto the reference area, or click to browse
* **Supported formats** — JPG, PNG, WebP
* **Auto-compression** — images are automatically compressed to max 2MB / 2048px before upload
* **Maximum source dimensions** — an image whose width or height exceeds **8000 pixels** is rejected before upload, with a message asking you to resize it. Everyday files stay well clear of this: a 4K photo is 3840px on its long edge, and an A4 scan at 300dpi is 2480×3508. Only print-scale source files (large-format banners, poster artwork at full print resolution) run into it — scale those down before uploading
* **Thumbnail grid** — uploaded images appear as thumbnails with an × button to remove each one
* **Per-model limits** — the maximum number of reference images varies by model (0–10; Seedream 5.0 Pro accepts the most at 10). Models that don't support references (Flux 2 Klein and Z Image Turbo) hide this section entirely.
See the [Models](/en/features/models) page for the exact reference image limit of each model.
### Adding References from the Gallery
You don't need to download images first. Drag any gallery card directly into the reference image area — the image URL is used automatically.
Combine gallery references with @ mentions for powerful multi-reference workflows. For example, drag two gallery cards as references, then write: "Combine the composition of @image1 with the color palette of @image2."
### Reference Types (Midjourney V8.1)
When using Midjourney V8.1 with a reference image, a dropdown appears to choose how the reference is interpreted:
| Type | Description |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Content Reference** | The AI uses the image as subject matter — it tries to recreate similar content. Control influence with the **Image Weight** (iw) parameter (0–2). |
| **Style Reference** | The AI extracts the visual style (colors, mood, technique) without copying the content. Control with **Style Weight** (sw, 0–1000). |
### When to Use Reference Images
Some prompts work best — or only make sense — with a reference image attached. MeiGen automatically detects keywords like "reference image" or "uploaded image" in your prompt and shows a hint if no reference is uploaded.
**Common patterns that need reference images:**
| Prompt Pattern | Why Reference Needed |
| ----------------------------------------------- | ------------------------------------------ |
| "Transform this reference image into..." | The AI needs a source image to transform |
| "Based on the uploaded image, create..." | Explicitly references an uploaded image |
| "Keep the composition of @image1 but change..." | Uses @ mention syntax to reference uploads |
| "Redraw / reimagine the reference photo in..." | Requires a source to redraw from |
If your prompt describes transforming, redrawing, or referencing a specific image, always upload that image as a reference first. Without it, the AI will generate something unrelated to your intent.
***
## Generation Options
### Count
Choose how many images to generate per request (1–4).
* **Free users** — limited to 1 image per request. A lock icon appears on the + button.
* **Paid users** — can generate up to 4 images in parallel per request.
### Aspect Ratio
The default is **Auto** — no need to pick manually, a suitable ratio is chosen for you. Drop into the dropdown to pick a specific ratio when you want precise control.
| Ratio | Orientation | Common Use |
| ------------------ | ------------- | ----------------------------------------------- |
| **Auto** (default) | — | Hands-off |
| 1:1 | Square | Social media profile, icons |
| 3:4, 4:5, 2:3 | Portrait | Phone wallpaper, posters |
| 9:16 | Tall portrait | Stories, Reels |
| 9:21 | Ultra tall | Vertical banners (GPT Image 2.0, Z Image Turbo) |
| 1:3 | Ultra tall | Vertical banners (GPT Image 2.0 only) |
| 1:4, 1:8 | Ultra tall | Scrolling banners (Nanobanana 2 only) |
| 4:3, 3:2 | Landscape | Desktop wallpaper |
| 16:9, 21:9 | Wide | Cinematic, ultrawide displays |
| 3:1 | Ultra wide | Header banners (GPT Image 2.0 only) |
| 4:1, 8:1 | Ultra wide | Panoramic banners (Nanobanana 2 only) |
Not all models support all ratios. The selector automatically filters to show only ratios supported by the current model. See the [Models](/en/features/models) page for each model's supported ratios.
### Resolution
The resolution options come from the model you picked — the selector only offers what that model supports:
| Model | Resolution options |
| ------------------------------------------------------- | ------------------------------ |
| Nanobanana 2, Nanobanana Pro, Seedream 4.5 | 2K / 4K (default 2K) |
| GPT Image 2.0 | 1K / 2K / 4K (default 2K) |
| Grok Imagine Quality, Midjourney V8.1, Seedream 5.0 Pro | 1K / 2K (default 1K) |
| Seedream 5.0 Lite | 2K / 3K |
| Flux 2 Klein, Z Image Turbo, Agnes Image 2.1 Flash | Fixed — no resolution selector |
**The highest option of each model is paid-only** and shows a lock icon for free users. On a 1K / 2K model that means 2K is locked; on a 2K / 4K model it's 4K; on a 2K / 3K model it's 3K. The same rule applies to 4k on video (Seedance 2.0 Pro and Veo 3.1).
***
## Midjourney V8.1 Advanced Options
When using Midjourney V8.1, the **Advanced Options** button lets you tune extra generation parameters. **Most users can leave everything at defaults.**
| Parameter | Range | Default | Description |
| -------------------------- | -------- | ------- | ---------------------------------------------------------------------------------------------------------------- |
| **Stylize** | 0–1000 | 100 | Controls how strongly Midjourney's aesthetic is applied. Lower = more literal, higher = more artistic. |
| **Chaos** | 0–100 | 5 | Adds variation between runs. Higher values produce more unexpected results. |
| **Raw Mode** | on/off | off | When on, produces images with a raw, less processed aesthetic; when off, the model's default styling is applied. |
| **Image Weight (iw)** | 0–2 | 1 | How much influence the content reference image has. Only visible when a content reference is set. |
| **Style Reference (sref)** | URL/text | — | Provide a URL to an image whose style you want to mimic. Disabled when reference images are set to Style mode. |
| **Style Weight (sw)** | 0–1000 | 100 | How strongly the style reference is applied. Only visible when sref or style reference images are present. |
***
## Describe Image
See an image you like but don't know how to describe it? **Describe Image** uses AI vision to analyze an image and generate a detailed text prompt that captures its content, style, and composition.
### How to Use
1. **Image detail dialog** — open any image's detail view and click the **Describe Image** button. The AI analyzes the image and generates a descriptive prompt.
2. **Drag and drop** — drag any external image (or a gallery card) onto the **Describe Image dropzone** above the prompt editor in the generation sidebar.
### What It Generates
The AI examines the image and produces a prompt describing:
* Subject matter and scene composition
* Art style and technique (photorealistic, illustration, watercolor, etc.)
* Lighting, color palette, and mood
* Camera angle and perspective
You can use the generated prompt as-is, or edit it to create variations of the original image.
Describe Image works well as a starting point. Combine it with [Prompt Enhancement](#prompt-enhancement) — first describe an image you like, then enhance the result for even more detail.
***
## Video Generation
MeiGen offers several video models — the current list and full specs are on the [Models](/en/features/models) page:
* **Seedance 2.0** — three tiers, per-second pricing, 4–15 seconds, supports reference-video continuation
* **Veo 3.1** — two tiers, per-generation pricing, 4 / 6 / 8 seconds, audio always on
* **Happyhorse 1.1** — per-second pricing, 3–15 seconds
* **Grok Video 1.5** — per-second pricing, 4–15 seconds, a first-frame image is required
* **Agnes Video 2.0** — per-second pricing, fixed 5 seconds, covered by daily free credits
Open the video sidebar from the **Generate Video** button in the floating dock on desktop, or from **Generate** in the bottom dock on mobile.
### Storyboard — Video Prompt Builder
Not sure how to write a video prompt? Click the **Storyboard** button in the video sidebar to turn a one-line idea into a complete, ready-to-shoot prompt. It's the video equivalent of image [Prompt Enhancement](#prompt-enhancement) — but instead of a single paragraph, it produces a **timed, shot-by-shot storyboard**.
Storyboard expands your brief (or an existing prompt) into two parts:
1. **Overview paragraph** — visual style, mood, the main subject, the environment, and the camera + lighting language.
2. **Shot-by-shot timeline** — a beat list from `0s` to your selected duration with no gaps, e.g. `0s-2s: slow push-in + ...`, including spoken lines or sound effects where they fit.
The timeline is built to match the **duration you've selected** in the sidebar, so set your duration first, then click Storyboard.
Storyboard keeps your input language — write your idea in any language and the storyboard comes back in the same language. It requires sign-in and shares a daily AI-assist limit with the other prompt tools, but does **not** spend image/video generation credits.
```text Input (one line) theme={null}
a barista making latte art in a cozy café
```
```text Storyboard (8s) theme={null}
A warm, photorealistic close-up film of a skilled barista crafting latte art
in a cozy sunlit café, soft morning light, shallow depth of field, gentle
handheld camera, an inviting and calm mood.
0s-2s: slow push-in on the barista's hands gripping the steaming pitcher
2s-5s: top-down shot as milk pours into the espresso, a white rosetta blooms
5s-8s: slow pull-back revealing the finished cup set on the counter
```
Keep your starting idea short — a sentence is enough. Storyboard works best when it has room to build the shots; if you already wrote a detailed timeline, you don't need it.
### Frame Images
Upload frame images to control the video's composition:
* **First Frame** — sets the starting composition of the video
* **Last Frame** — sets the ending composition
**Seedance 2.0** and **Veo 3.1** accept both frames (up to 2). **Happyhorse 1.1**, **Grok Video 1.5** and **Agnes Video 2.0** accept a first frame only.
Frames are optional except on **Grok Video 1.5**, which is image-to-video only — a first frame is required, and the output ratio follows that image.
You can also drag gallery cards directly as frame images.
### Seedance 2.0
| Option | Values | Description |
| ------------------- | ------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------- |
| **Aspect ratio** | adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 | 7 ratios. `adaptive` matches the reference image/video dimensions |
| **Resolution** | Mini / Fast: 480p, 720p. Pro: 480p, 720p, 1080p, 4K (default 480p) | Depends on the selected tier; 480p saves credits |
| **Duration** | 4–15 seconds (default 4) | Slider or input |
| **Audio** | Auto-generated | Always on |
| **Reference video** | Yes | Used as visual conditioning input for "video continuation" (see the [Extending a Video](#extending-a-video) section below) |
**Pricing**: per-second; full formula (including the minimum-billable table for continuation) is in the [Models](/en/features/models) page. The sidebar shows the exact total before you submit.
**Three modes**:
* Text-to-video: just write a prompt
* Image-to-video: upload 1–2 frame images to control first/last frame
* Video extension: upload a reference video (see the [Extending a Video](#extending-a-video) section below)
**Adding a reference video makes the total credit cost depend on its length**: billed seconds = "reference video duration + your chosen duration" (never below the minimum-billable floor). Longer reference videos cost more. The sidebar shows the exact total before you submit — confirm before clicking Generate.
### Veo 3.1 — Two Tiers, Native Audio
| Option | Values | Description |
| ---------------- | --------------------------------- | ---------------------------------------------------------------------- |
| **Tier** | Fast (default) / Pro | Toggle via the segmented control under the model picker |
| **Aspect ratio** | Auto, 16:9, 9:16 | Landscape or portrait only; Auto infers from prompt or reference image |
| **Resolution** | 720p / 1080p / 4k (default 1080p) | All three share the same price; 4k just renders longer |
| **Duration** | 4 / 6 / 8 seconds (default 4) | Pick from the dropdown |
| **Audio** | Auto-generated | Always on |
**Pricing**: per generation, depends on tier × duration. Full table is in [Models → Veo 3.1](/en/features/models).
Full specs and a side-by-side comparison of all video models are on the [Models](/en/features/models) page.
### Converting Images to Video
After generating an image, hover over the completed card and click the **Animate** button. This sends your generated image to the video sidebar as the first frame, making it easy to turn a still image into a short video clip.
### Extending a Video
Seedance 2.0 only. Upload an existing video as visual conditioning input to guide the model toward generating a new segment that semantically continues the original.
#### Three entry points
1. **From a video card**: hover over any completed video card and click the **Extend** button in the top-left corner — the video loads into the Seedance sidebar as a reference
2. **Drag a video to the reference-image area**: drag a video (local or from the gallery) onto the reference upload area — a confirmation popup asks "Use this as an extension reference?"
3. **Direct upload**: in the Seedance sidebar's "Extend or reference video" upload area, pick a local video
#### Workflow
1. Trigger any of the entries above — the reference video appears in the sidebar preview
2. The prompt box is auto-prefilled with `"Extend this video with the following plot: "` — keep this prefix and add your desired continuation
3. Pick resolution and your new video's duration (4–15 seconds)
4. Click Generate
#### Key behavior
**Your prompt must explicitly say to extend the video.** If the prompt only describes a scene without mentioning extension, the new segment may drift away from what you imagined as a "continuation" (the visuals will still be influenced by the reference, but the plot direction follows whatever your prompt says). The auto-prefilled `Extend this video with the following plot:` prefix exists to nudge the model toward "continuation" semantics — keep it and append your content.
**Output length = your selected duration** (4–15 seconds), independent of the reference video's length. The reference video acts as visual conditioning input (influencing the new segment's framing, style, characters, etc.) and is not concatenated into the output. If you want a continuous "original + new" video, you'll need to stitch them locally with a video editor.
#### Pricing
Extension uses a separate With-reference-video rate table (Mini 480p 7 credits/sec, 720p 13 credits/sec; Fast 480p 8, 720p 14; Pro 720p 16, 1080p 28). **Billable seconds = reference video duration + your selected duration** (not less than the minimum-billable floor). The full formula, lookup table, and examples are in [Models → Seedance 2.0](/en/features/models).
Typical costs (examples):
* Upload a 3-second reference + generate 5 seconds at **Fast** 480p: **72 credits** (billable = max(3+5, 9) = 9 sec × 8)
* Upload a 5-second reference + generate 10 seconds at **Fast** 720p: **238 credits** (billable = max(5+10, 17) = 17 sec × 14)
***
## Generation History
After submitting a generation request, your images appear as cards in the generation history panel (accessible via the floating dock's History button or the sidebar).
### Pending State
Cards show live progress while generating; video generations also display an estimated completion time (e.g., `2–5 min`).
Typical generation times by model:
| Model | Typical Time |
| ----------------------- | ------------------------------------------------------------------------------------------------------------------------ |
| Grok Imagine Quality | \~9 seconds (1K) / \~17 seconds (2K) |
| Z Image Turbo | \~11 seconds |
| Nanobanana 2 | \~20 seconds |
| Flux 2 Klein | \~21 seconds |
| Nanobanana Pro | \~30 seconds |
| Agnes Image 2.1 Flash | \~30 seconds |
| Seedream 4.5 | \~32 seconds |
| GPT Image 2.0 | \~35 seconds |
| Seedream 5.0 Lite | \~42 seconds |
| Midjourney V8.1 | \~50 seconds |
| Seedream 5.0 Pro | \~2 min (1K) / \~2.5 min (2K) |
| Grok Video 1.5 (video) | \~35 seconds |
| Agnes Video 2.0 (video) | \~2 min |
| Seedance 2.0 (video) | \~2–3 min (Mini, default) / \~2–4 min (Fast) / \~3–6 min (Pro, longer at 4K) |
| Happyhorse 1.1 (video) | \~2–4 min (720p) — varies with clip length, and runs faster when you supply a first frame; 1080p takes noticeably longer |
| Veo 3.1 (video) | \~1–4 min (720p / 1080p) / \~6–8 min (4k) |
### Failed State
If a generation fails, the card shows:
* The reason for failure
* **"Credits refunded"** note — credits are automatically returned when a generation fails
* **Retry button** — click to resubmit the same request with the original prompt and settings
### Completed State
Hover over a completed generation card to reveal action buttons:
| Action | Description |
| -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Edit** | Adds the generated image as a reference for a new generation |
| **Cutout** | Removes the background, creating a transparent PNG (useful for logos, stickers, product compositing). Creates a new card; the original is preserved |
| **Animate** | Converts the still image into a video (image-to-video) |
| **Delete** | Removes the card from your history |
| **Use Idea** | Copies the prompt back to the editor for reuse |
| **Save** | Saves to your favorites collection |
| **Download** | Saves the image to your device |
| **Share to X** | Opens a pre-filled tweet with your generated image |
The card also shows the aspect ratio and resolution labels in the bottom-left corner.
***
## Credits and Billing
MeiGen uses a dual credit system:
* **Daily credits**: a free daily allowance that resets at 00:00 UTC, usable on **basic models** — currently **Z Image Turbo**, **Flux 2 Klein** and **Agnes Image 2.1 Flash** for images, and **Agnes Video 2.0** for video (the only video model daily credits cover). They are grouped under **Basic Models** in the model picker.
* **Purchased credits**: your non-expiring balance, shown as **Credits** in the sidebar. It covers everything outside the daily allowance — credits you topped up, plus any bonus credits granted to your account (sign-up welcome bonus, referral rewards). Usable on all models.
Generating with premium models (GPT Image 2.0, Nanobanana 2 / Pro, Seedream, Midjourney V8.1, Grok Imagine Quality, Seedance 2.0, Veo 3.1, Happyhorse 1.1, Grok Video 1.5, and so on) requires welcome or purchased credits — daily credits never apply to them, whether or not you have purchased before.
**API Token calls** only use purchased credits, regardless of the model; daily credits never apply to the API path.
The credit cost is shown on the Generate button before you submit. Hover over the credits card at the bottom of the left sidebar to see your balance breakdown.
### Free vs Paid
| Feature | Free | Paid |
| -------------------------------- | ---------------------------------- | ---------------------------------- |
| Highest resolution of each model | Locked | Available |
| Images per request (Web UI) | 1 | Up to 4 |
| API access | Available (purchased credits only) | Available (purchased credits only) |
> The "Images per request" limit applies to the Web UI only. Each API call always returns 1 image (Midjourney V8.1 always returns 4 candidates per call).
>
> Creating an API token does not require a paid account — any account can issue one from Settings → API Keys. What API calls do require is purchased credits, since daily free credits never apply on the API path.
***
## FAQ
Generation times vary by model and server load. See the table at the top of this section for typical times.
Credits are automatically refunded when a generation fails. The refund is reflected immediately in your balance. You can use the Retry button to try again.
Use the **Retry** button on a failed card, or click **Use Idea** on a completed card to copy its prompt back to the editor. Note that AI generation is non-deterministic — even with the same prompt, results will vary.
**Cutout** removes the background from a still image, creating a transparent PNG. **Animate** converts the still image into a short video clip using the video generation model.
The top resolution of each model is reserved for paid users — 4K on models offering 2K / 4K, 3K on Seedream 5.0 Lite, and 2K on models offering 1K / 2K. Upgrade your account to unlock it.