> ## Documentation Index
> Fetch the complete documentation index at: https://aspect.inc/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Choose an AI generation model

> Aspect's recommended models for initial images, precise edits, creative graphics, continuous takes, scene changes, voice, and music.

**Last verified: November 5, 2026.** Model versions and input limits below are those available in Aspect on this date.

## Recommended workflow

1. **Plan the shot with hand sketches.** Work out the framing, subject placement, camera movement, and any cuts before generating images.
2. **Get the right look with image generation.** Use Nano Banana 2.1 for initial shot images. Try Sunburst or Krea 3 for more expressive graphics and backgrounds. Use FLUX 3 image for small edits, then settle on the images you want to animate.
3. **Choose the video model for the shot.** Use Kling 4 for a single continuous take. Use Seedance 2.5 for cuts, scene changes, people, or several references.
4. **Generate at low resolution first.** Check the movement, timing, and transitions before spending more on a higher-resolution version. This is also the recommended starting point for students learning the workflow.
5. **Regenerate at higher resolution once you are happy.** Use the approved images and shot plan for the final run. Review it again, because a new generation can change details as well as resolution.

## Pick a model

**Seedance 2.5 is our best overall video model.** It is especially good for videos with cuts and scene changes, shots with people, and work that needs several references. **Kling 4** is our other main video recommendation: choose it for a single continuous take, including product reveals and controlled camera moves.

For images, start with **Nano Banana 2.1** to establish the look. Use **FLUX 3 image** for small edits to an image you already like.

For graphics such as flyers and posters, our pick is **GPT Image 2.5 Sunburst**. Use **GPT Image 2.5 Flare** when you want to explore concepts quickly. Sunburst emphasizes editing precision; Flare emphasizes speed. Both can generate and edit graphics.

**Krea 3 and Sunburst are our most creative image choices.** Try them for atmospheric graphics, backgrounds, and a more expressive visual style with a feel similar to Midjourney.

| Your job | Recommended model | Why |
| - | - | - |
| Initial images, product stills, or storyboard frames | **Nano Banana 2.1** | Our first choice for getting the composition and visual style right. |
| Small edits, color versions, or background swaps | **FLUX 3 image** | Our best image-editing choice when the existing image is already close. |
| Flyers, posters, or graphics with text | **GPT Image 2.5 Sunburst** | Our choice for finished graphics. Check the wording and placement. |
| Fast graphic concepts and design variations | **GPT Image 2.5 Flare** | The speed-oriented option for trying several designs. |
| Atmospheric graphics and creative backgrounds | **Krea 3 or GPT Image 2.5 Sunburst** | Our most creative picks for expressive styling and mood. |
| A video with cuts and scene changes | **Seedance 2.5** | Our best overall video model, especially for sequences that move between shots. |
| A single continuous take or product reveal | **Kling 4** | Our preferred model for keeping one shot going without cuts. |
| Video guided by several image, video, or audio references | **Seedance 2.5** | Its reference mode accepts all three types of media. |
| People, acting, or faces in a scene | **Seedance 2.5** | Our preferred video model for people. Real-face input restrictions can apply. |
| A talking head performing an exact recording | **Kling AI Avatar 2 Pro** | Takes a portrait and recorded speech, so the recording controls the words and timing. |
| Spoken narration | **ElevenLabs TTS 4** | A dedicated text-to-speech model for expressive delivery. |
| Music cues or songs | **ElevenLabs Music 2.5** | Generates music from a written brief. |
| Standalone sounds or atmosphere | **Stable Audio 3** | Generates audio without requiring a video. |
| Sound for a finished video | **MMAudio 2** | Uses the video to guide the sound. |

## Model reference

These inputs and limits describe the controls available in Aspect. A provider's own app may offer different options. **Reference images** guide the subject, composition, or style. **First and last frames** specify the opening and ending images of a video. Resolution labels such as 1K and 4K do not guarantee the same pixel dimensions or amount of detail across providers.

### Recommended image models

> [accordion] Nano Banana 2.1 — best for initial images
>
> **Best for:** establishing the look of a shot, product and e-commerce stills, storyboard frames, and reference-led compositions. Start here when you need a strong image to animate later.
>
> **Inputs and limits:** text and up to 14 reference images; 1K, 2K, or 4K output.
>
> **Watch for:** labels, faces, and details that change between versions. For a small correction to an otherwise approved image, use FLUX 3 image.

> [accordion] GPT Image 2.5 Sunburst — graphics and creative visual direction
>
> **Best for:** finished flyers, posters, layouts, and atmospheric backgrounds. Alongside Krea 3, it is one of our most creative image choices for expressive graphics.
>
> **Inputs and limits:** text and up to 16 reference images; 1K, 2K, or 4K output. Effort settings control the generation quality.
>
> **Sunburst or Flare:** choose Sunburst when editing precision matters. Choose Flare when you want to explore several concepts quickly. Both can make graphics; Sunburst is our preferred starting point for the finished design.
>
> **Watch for:** spelling, exact text placement, and logos. Check the finished graphic at its intended size before using it.

> [accordion] GPT Image 2.5 Flare — fast graphic concepts
>
> **Best for:** quickly exploring layouts, graphic concepts, and image variations before choosing a direction.
>
> **Inputs and limits:** text and up to 16 reference images; 1K, 2K, or 4K output. It offers the same effort settings as Sunburst.
>
> **Watch for:** speed does not establish which model will handle a particular font, layout, or brief better. Check the result against the design requirements.

> [accordion] Krea 3 — creative graphics and backgrounds
>
> **Best for:** atmospheric backgrounds, expressive graphics, and exploring a visual mood. Alongside Sunburst, it is our pick for a more stylized feel similar to Midjourney.
>
> **Inputs and limits:** text only; configured at 1K in Aspect.
>
> **Watch for:** it cannot take a sketch, product photo, or existing image as a reference in Aspect. Use Sunburst when this kind of creative direction needs reference images or a higher-resolution output.

> [accordion] FLUX 3 image — best for small edits
>
> **Best for:** small changes to an existing image, including color versions, background swaps, and adjustments to a product or subject. Supply the existing image as the edit input.
>
> **Inputs and limits:** FLUX 3 image and FLUX 3 Klein accept text and up to 10 reference images. Both are configured at 1K in Aspect.
>
> **Watch for:** changes outside the requested area. Inspect the whole image after an edit. The FLUX 3 recommendation here refers to the image model; FLUX video models are separate.

### Recommended video models

> [accordion] Seedance 2.5 — best overall video model
>
> **Best for:** videos with cuts and scene changes, people and performances, and shots that combine several references. Choose it when the sequence needs to move between scenes or draw on multiple pieces of visual guidance.
>
> **Inputs and limits:** requires a visual input. Use first/last frames **or** reference mode. Reference mode accepts up to 30 images, 10 videos, and 10 audio clips, with 50 items total. Output is 4–30 seconds at 480p, 720p, or 1080p.
>
> **Watch for:** first/last frames and reference mode cannot be combined in the same run. Real-face uploads have additional restrictions. Several references guide the result but do not guarantee that every detail will remain identical across shots.

> [accordion] Kling 4 — best for a single continuous take
>
> **Best for:** uninterrupted shots, product reveals, and camera moves that stay within one scene. Kling 4 and Seedance 2.5 are our two main video recommendations.
>
> **Inputs and limits:** text alone, first/last frames, and up to eight reference images; 3–30 seconds at 720p or 1080p.
>
> **Watch for:** faces and people are less dependable than in our Seedance work. Aspect does not expose video references or middle keyframes for Kling 4. Choose Seedance for sequences with cuts or several kinds of reference media.

> [accordion] Kling AI Avatar 2 Pro and Standard — talking heads from recorded speech
>
> **Best for:** a presenter or character performing an existing recording. This is a separate model from Kling 4.
>
> **Inputs and limits:** a portrait first frame and audio are required. Audio can be 2–300 seconds; output follows the recording. Aspect does not declare a resolution limit for these entries.
>
> **Watch for:** this is a talking-head workflow rather than a general action or camera-move model. The recording determines pronunciation and pauses.

> [accordion] Veo 3.1 and Fast — short scenes with generated sound
>
> **Useful for:** short scenes combining picture, dialogue, and atmosphere.
>
> **Inputs and limits:** text and first/last frames; 4, 6, or 8 seconds. Output can reach 4K, depending on the supported combination of settings. Audio is optional.
>
> **Watch for:** the eight-second limit per run. Reference-image and extension features offered elsewhere are not exposed by these Aspect entries.

### Audio and voice

> [accordion] ElevenLabs TTS 4 and TTS Turbo 4 — spoken narration
>
> **Best for:** turning a script into spoken narration. TTS means text-to-speech.
>
> **Inputs and limits:** text only, up to 5,000 characters in total, including text from connected nodes, transcripts, and documents. Aspect declares no fixed output duration. Turbo 4 is another available variant.
>
> **Watch for:** pronunciation, acronyms, and names. These nodes do not accept a recording as a voice-cloning reference.

> [accordion] ElevenLabs Music 2.5 — music cues and songs
>
> **Best for:** music with a planned mood and structure.
>
> **Inputs and limits:** text; 30, 60, 120, 180, or 300 seconds.
>
> **Watch for:** prompts naming artists, songs, or albums, and prompts containing distinctive copied lyrics. These are restricted by the provider.

> [accordion] Stable Audio 3 and Remix — sounds and atmosphere
>
> **Best for:** standalone sounds, atmosphere, and music experiments. Remix starts from an existing recording.
>
> **Inputs and limits:** text for generation; text plus audio for Remix. Generation offers 10, 30, 60, 90, or 180 seconds. Remix offers 30, 60, or 180 seconds.
>
> **Watch for:** exact scripted speech and effects that must land on a precise video frame. These nodes do not use a video to control timing.

> [accordion] MMAudio 2 — sound for an existing video
>
> **Best for:** generating sound guided by a finished shot.
>
> **Inputs and limits:** video required, with text guidance. Sound is generated for at most 30 seconds of the video.
>
> **Watch for:** dialogue accuracy and the timing of individual effects. Do not assume the output will include separately editable sound stems.

> [accordion] ElevenLabs Voice Changer — a new voice for an existing performance
>
> **Useful for:** changing the voice of an existing recording.
>
> **Inputs and limits:** audio required, up to 300 seconds; output duration follows the input. This catalog entry has no version number.
>
> **Watch for:** it takes an audio performance, not a written script. Use a recording and voice you are authorized to use.

> [accordion] Gemini 3.8 Flash TTS — another narration option
>
> **Useful for:** trying an alternative speech delivery.
>
> **Inputs and limits:** text only. Aspect declares no duration or audio-format ceiling.
>
> **Watch for:** pronunciation and pacing. Audio-reference and voice-cloning controls are not available through this entry.

> [accordion] Lyria 3.5 — another music option
>
> **Useful for:** music with sections such as verses, choruses, and bridges.
>
> **Inputs and limits:** text only; the provider lists 44.1 kHz stereo output. Aspect declares no selectable duration or maximum.
>
> **Watch for:** the returned length. Check it before building an edit around the cue.

> [accordion] CassetteAI Music Generator — background-music sketches
>
> **Useful for:** exploring background music for an edit.
>
> **Inputs and limits:** text; 30, 60, 120, or 180 seconds. The provider lists 44.1 kHz stereo output. This catalog entry has no version number.
>
> **Watch for:** whether the rhythm and arrangement fit the picture.

### Other available image models

> [accordion] Seedream 5 Pro and Flash
>
> **Useful for:** trying alternative designed compositions and image variations.
>
> **Inputs and limits:** text and up to 10 reference images; configured at 1K.
>
> **Watch for:** small typography and unintended changes to areas you wanted to preserve.

> [accordion] Recraft 4.1
>
> **Useful for:** illustration and visual-style exploration.
>
> **Inputs and limits:** text only; configured at 1K.
>
> **Watch for:** this entry produces raster images, not editable vectors. It does not accept an existing product photo for editing.

> [accordion] Qwen Image 3
>
> **Useful for:** trying image changes with a small reference set.
>
> **Inputs and limits:** text and up to three reference images; configured at 1K.
>
> **Watch for:** small text and details that shift during an edit.

> [accordion] Grok Imagine 2.0
>
> **Useful for:** alternative interpretations of a text brief.
>
> **Inputs and limits:** text only; 1K or 2K.
>
> **Watch for:** this entry cannot take an existing image for a background swap or product variation.

> [accordion] Muse Image
>
> **Useful for:** experimenting with compositions guided by several references.
>
> **Inputs and limits:** text and up to 10 reference images. Aspect does not declare an output-resolution limit.
>
> **Watch for:** consistency of product details, faces, and text between results.

> [accordion] Ideogram 3
>
> **Useful for:** trying an alternative text-led graphic or layout.
>
> **Inputs and limits:** text and one image reference for remix; configured at 1K.
>
> **Watch for:** exact wording and placement. Check every line of text before delivery.

### Other available video models

> [accordion] LTX 2.5 Pro and Fast
>
> **Useful for:** short-shot experiments and scenes built around existing audio.
>
> **Inputs and limits:** the normal Pro and Fast entries accept text and first/last frames, with 6, 8, or 10-second output. Pro reaches 1080p; Fast offers up to 2160p. Separate **Audio to Video** entries require audio and optionally accept a first frame. Pro accepts 2–10 seconds and Fast accepts 2–20 seconds, both at 1080p.
>
> **Watch for:** choose the Audio to Video entry when a recording needs to control the timing.

> [accordion] H3 Max and H3 Max Turbo
>
> **Useful for:** trying another interpretation of a short shot.
>
> **Inputs and limits:** text, first/last frames, and one audio reference; 1–15 seconds at 480p, 768p, or 1080p.
>
> **Watch for:** do not assume Max and Turbo will handle the same shot equally well.

> [accordion] Mischief 1
>
> **Useful for:** short motion experiments from a still image.
>
> **Inputs and limits:** first frame required; 3 or 5 seconds at 480p or 720p.
>
> **Watch for:** it cannot generate from text alone and does not support longer takes.

> [accordion] Grok Imagine Video 1.5 and Lite
>
> **Useful for:** generating short clips from text alone.
>
> **Inputs and limits:** text only in Aspect; 1–15 seconds, up to 1080p. Lite's 1080p output is upscaled from 720p.
>
> **Watch for:** neither Aspect entry accepts your photo as an input.

> [accordion] Gemini Omni Flash 1.1
>
> **Useful for:** short experiments with opening and ending frames.
>
> **Inputs and limits:** text and first/last frames; 3–10 seconds, with resolution options from 360p to 4K.
>
> **Watch for:** a higher output resolution does not guarantee better motion or more consistent faces.

> [accordion] Wan 3.0 and Prime
>
> **Useful for:** trying longer uninterrupted shots.
>
> **Inputs and limits:** text and first/last frames; 2–30 seconds, up to 1080p.
>
> **Watch for:** Aspect's entries do not expose the broader reference and editing controls available in some provider workflows.

> [accordion] FLUX 3 Video and Max Video
>
> **Useful for:** exploring animation and motion-design looks.
>
> **Inputs and limits:** text and first/last frames; 5–20 seconds, up to 1080p.
>
> **Watch for:** these are separate from FLUX 3 image. The image-editing recommendation above does not apply to them.

> [accordion] PixVerse 6
>
> **Useful for:** short-shot experiments with opening and ending frames.
>
> **Inputs and limits:** text and first/last frames; 1–15 seconds, up to 1080p. Audio is optional.
>
> **Watch for:** this entry does not expose reference-video input.

## Content restrictions

* **Brands and logos:** a prompt or reference can be rejected by a provider's content filters. Having permission to use an asset does not guarantee the filter will accept it.
* **Real people:** Seedance 2.5 restricts ordinary direct uploads containing real human faces. Approved portrait workflows are not confirmed in Aspect. Veo also applies person and celebrity filters.
* **Voices and music:** use voices you are authorized to use. ElevenLabs Music restricts artist names, song and album titles, and distinctive copied lyrics in prompts.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.