# Video Workflows
import { Card, CardGrid, Steps, LinkButton } from '@astrojs/starlight/components';
InvokeAI ships **twenty-six ready-made video workflows** — ten for **Wan 2.2**, six for **MiniMax H3** and ten for **LTX-2** — so you can start generating video without wiring up nodes yourself. Each one is a complete recipe — open it from the **Workflows** library, pick your models, type a prompt (and/or drop in an image), and press **Invoke**.
This page helps you choose the right workflow and run it. For how the models work under the hood, VRAM details, and troubleshooting, see the [Video Generation guide](/users-guide/video-generation/introduction/).
:::tip[There's an easier way]
Most of what these workflows do is also available with no graph at all: the **Video panel** in the editor's left rail infers the right pipeline from whatever you fill in — prompt, first/last frame, initial video — and handles model constraints automatically. Start with the [Video Generation guide](/users-guide/video-generation/introduction/) if you just want to make a video; come here when you want graph-level control.
:::
:::caution[Experimental]
Video generation is a prototype feature. These workflows and the models behind them are new — expect rough edges, and check the [Video Generation guide](/users-guide/video-generation/introduction/) if something misbehaves.
:::
---
## Before you start
**1. Install the video models.** Open the **Model Manager** and install a starter bundle: **Wan 2.2**, **MiniMax H3**, and/or **LTX 2.5**. The [Video Generation guide explains the bundles and what fits your GPU](/users-guide/video-generation/introduction/#the-models).
**2. You'll select models when you open a workflow.** The shipped workflows come with their model slots **left empty on purpose** — every InvokeAI install stores models a little differently, so a workflow can't point at yours automatically. When you open one, look at its **Notes** panel (the note icon in the workflow editor): it lists exactly which model to pick in each slot. Drop those in once and you're set.
:::tip
Whenever a workflow won't run because a model field is empty, open its **Notes** — the shopping list of models to select is right there.
:::
---
## Which workflow should I use?
The Wan workflows come in three families — **Text to Video**, **Image to Video**, and **Extend Video** — each with a high-quality option, a "concept LoRA" variant, and a lighter low-VRAM option — plus an **Interpolate** workflow that bridges two images and an **Extend Video to Image** variant that steers a continuation toward a target frame. The six **MiniMax H3** workflows cover the same conditioning modes — plus reference-conditioned generation — and add sound. The ten **LTX-2** workflows cover every one of those modes with sound, add **audio-to-video** and **video-to-audio**, and include a two-stage high-resolution text-to-video and a LoRA-collection variant.
| I want to… | Use | Notes |
|---|---|---|
| Make a clip from a text description | **Text to Video - Wan 2.2 Lightning** | Best quality, fast. Start here. |
| Make a clip **with a soundtrack** | **Text to Video - LTX-2** or **Text to Video - MiniMax H3** | Both generate synchronized stereo audio with every video. |
| Make a **high-resolution** clip | **Text to Video - LTX-2 Two-Stage** | Generates at half size, upscales, then refines at 1024p or 1536p. Slow. |
| Animate an existing image | **Image to Video - Wan 2.2 Lightning**, **First Frame to Video - MiniMax H3** or **First Frame to Video - LTX-2** | Your image becomes the first frame. |
| Make a video that *ends on* an image | **Last Frame to Video - MiniMax H3** or **Last Frame to Video - LTX-2** | The video builds toward your image. |
| Make a video between two images | **Interpolate 2 Images to Video - Wan 2.2 Lightning**, **First and Last Frame to Video - MiniMax H3** or **First and Last Frame to Video - LTX-2** | Provide a start and end image; the model interpolates between them. |
| Make a video longer than one generation | **Extend Video - Wan 2.2 Lightning**, **Extend Video - MiniMax H3** or **Extend Video - LTX-2** | Continues an existing video and stitches the pieces together. H3 and LTX-2 continue the soundtrack too. |
| Make a clip **in the style/likeness of reference media** | **Reference to Video - MiniMax H3** | Conditions on up to 3 reference videos and 9 images (Ref2VA transformer required); reference order changes the result. |
| Extend a video toward a specific end frame | **Extend Video to Image - Wan 2.2 Lightning** or **Extend Video to Image - LTX-2** | Continues a video and interpolates the new part to a target image. |
| Make a picture **for a soundtrack** — a song, a voice recording | **Audio to Video - LTX-2** | The audio is kept as-is; the video is generated to match it, including lip movement. |
| Add a **soundtrack to a silent or existing clip** | **Video to Audio - LTX-2** | Your footage is kept as-is; a new stereo soundtrack is generated for it. |
| Add my own style/subject LoRAs | the **w/ Concept LoRAs** variant of a Wan workflow, or **Text to Video - LTX-2 w/ LoRA Collection** | Same as the base workflow, plus your LoRAs. |
| Run on a smaller GPU (≈12–16 GB) | a **TI2V-5B (Low Quality)** workflow | Smaller model, lower memory, lower quality. |
If you're not sure, start with **Text to Video - Wan 2.2 Lightning** to get a feel for it, then move to Image-to-Video and Extend — or straight to **Text to Video - MiniMax H3** if you want audio and have the hardware for it.
---
## Bundled Wan 2.2 workflows
### Text to Video
Type a prompt, get a ~5-second clip. Uses the high-quality A14B model with the Lightning speed-up (just 4 steps). **The recommended starting point.**
The same workflow, plus extra slots for your own concept LoRAs (a style or character you've trained or downloaded). Every LoRA slot must be filled — if you don't have concept LoRAs to add, use the plain version instead.
Uses the smaller TI2V-5B model. Lower quality and slower per result, but fits comfortably on 12–16 GB GPUs. Good for quick drafts or smaller cards.
### Image to Video
Drop in a starting image and a prompt; the model animates outward from your image as the first frame. Uses the A14B model + Lightning. **The recommended image-to-video starting point.**
Same as above, with slots for your own concept LoRAs.
Image-to-video on the smaller TI2V-5B model. Lower quality, lower VRAM — the low-memory option for animating an image.
### Interpolate Between Two Images
Provide a **start image** and an **end image**; the model generates a clip that begins on the first and animates smoothly to the second (first-last-frame interpolation). Great for morphing between two stills or bridging two shots. Uses I2V-A14B + Lightning (4-step).
:::caution[The "Concept LoRAs" variants need a LoRA in every slot]
A **concept LoRA** is an add-on that teaches the model a specific style, character, or subject. The *w/ Concept LoRAs* workflows are the same as their plain counterparts but add extra LoRA slots. **Every LoRA slot is required** — if any slot, including the extra concept ones, is left empty, the workflow won't run. Only use these variants when you actually have concept LoRAs to apply; otherwise use the plain version.
:::
:::note[About the "TI2V-5B (Low Quality)" variants]
"Low quality" here means *lower than the A14B Lightning workflows*, in exchange for running on much less video memory. On a 12–16 GB GPU they may be your best (or only) option, and they're handy everywhere for fast drafts.
:::
### Extend Video
The Wan models are trained on short (~5 second) clips, so the way to make something longer is to **continue** a video and stitch the pieces together. These workflows do that for you.
Provide an existing video and use the sliders to choose where to continue from. The workflow generates a new clip starting from that point and joins it onto the original with a smooth transition. Run it repeatedly to keep growing the video.
Same as above, with slots for your own concept LoRAs so the continuation keeps your style or subject.
Continue a video **toward a target image**. Provide a starting video and a destination image; the new segment interpolates from the video's last frame to your image (FLF2V), then joins onto the original with a smooth transition. Combines the Extend and Interpolate ideas — useful for steering where a continuation ends up.
:::tip[Using "Extend Video to Image"]
This variant takes **two inputs** — the source video and a destination image:
1. Drop your **video** into the video input and use the **start/end frame sliders** to pick the slice to continue from.
2. Drop a **destination image** into the image input. The generated continuation interpolates from the video's last selected frame to this image, so the extended clip *ends on your image*.
3. **Invoke.** The new segment is concatenated onto the source with a cross-fade — tune the join (transition type and frame count) on the **Concatenate Videos** node.
Because the endpoint is pinned to your image, it's ideal for bridging a clip to a known next shot or landing on a specific final frame.
:::
The **Concatenate Videos** node smart-renders compatible H.264 MP4 clips by preserving compressed video outside transition regions. Clips must share dimensions and frame rate and use compatible encoding settings. Other inputs, or a failed smart-render validation, automatically use the regular full-encoding path. Set the node's **Smart Render** input to `false` to always use full encoding.
---
## Bundled MiniMax H3 workflows
One model family, six workflows, every conditioning mode — and every output is an MP4 with a **jointly generated stereo soundtrack**. All default to a **Turbo** step-distillation LoRA (6 steps for FL2VA with the original Turbo LoRA, 8 with either LightX2V release); H3 has no CFG or negative prompt to tune, and always renders at 24 fps. **Reference to Video** needs the Ref2VA transformer selected in the model loader; the other five use FL2VA.
Type a prompt, get a clip **with sound**. H3's vision-language text encoder follows detailed scene descriptions well — describe the audio you expect, too.
Drop in a starting image; the video animates outward from it, soundtrack included.
The reverse — provide an **ending** image and the video builds toward it. Unique to H3; no Wan workflow can do this.
Provide both; the model interpolates a clip (with audio) that starts on the first image and lands on the second.
Continue an existing video with a new H3 segment, joined on with a smooth transition. The continuation renders at H3's fixed 24 fps.
:::note[Selecting the H3 models]
The H3 workflows use the **MiniMax H3 Model Loader**, which takes three selections from the **MiniMax H3** starter bundle: the **MiniMax H3 Components** install as the main model, plus the single-file **FL2VA Transformer (int8, pruned)** and **Text Encoder (int8)** in its override fields. Each workflow's **Notes** panel lists them. H3 is the heaviest local video option — see [the hardware notes](/users-guide/video-generation/minimax-h3/#installing-minimax-h3).
:::
:::caution[License restrictions]
MiniMax H3 is distributed under a restrictive community license that excludes certain territories (currently the US, EU, UK, and South Korea) and extends to outputs. See the [Video Generation guide's license note](/users-guide/video-generation/minimax-h3/#installing-minimax-h3) before installing.
:::
---
## Bundled LTX-2 workflows
Every LTX-2 workflow outputs an MP4 with a **jointly generated stereo soundtrack** (or, for audio-to-video, your own). Each works with either transformer: the **Dev** build runs the guided 30-step recipe, and the **Distilled** build a fixed 8 steps that ignore the Steps and CFG fields. The canvas comes from your image, source video or clip where there is one — except in Audio to Video, which generates the picture — and otherwise the **Width** and **Height** fields set only the aspect ratio. **Target Resolution** sets the size.
Type a prompt, get a ~5-second clip with sound. Describe the audio you expect, not just the picture.
The high-resolution version: generates at half the canvas, doubles it with LTX-2's latent upscaler, then refines it at full size. Defaults to 1024p (1792×1024 at 16:9); 1536p also works, but 512p–768p fail here — use the plain Text to Video workflow for those. Much slower — minutes on Distilled, far longer on Dev.
Text-to-video plus a list of your own LTX-2 LoRAs, each with its own weight. Leave the list empty to run the base model.
Your image becomes the opening frame; the video and its soundtrack continue from it.
Provide an **ending** image and the video builds toward it.
Provide both; the clip starts on the first image and lands on the second. Give the two images the same aspect ratio.
Continue an existing clip. The continuation starts from the source's last few frames **and its closing sound**, and joins onto it with a cross-fade over the same span, so the picture and the soundtrack both carry across the join. **Context Frames** sets how much of the source is carried in.
The same, with a **Last Frame** image: the continuation ends on your image. Give it the source clip's aspect ratio.
Drop in a clip — or an audio file uploaded to the gallery — and LTX-2 generates a picture for its soundtrack, down to lip movement. The soundtrack decides the length, and the output carries the original recording. Say in the prompt who or what is making the sound.
Drop in a clip and LTX-2 generates a soundtrack for it. The clip decides the length and frame rate, and the output is your original footage with the new audio.
:::tip[Tuning the LTX-2 workflows]
* **Use a single-pass resolution** (512p, 704p or 768p) everywhere except the Two-Stage workflow. The other workflows have no refine pass, so 1024p and 1536p render *smaller* than 768p. Audio-to-video and video-to-audio cannot run two-stage at all.
* **Number of Frames** must be on LTX-2's 8·n + 1 grid (9, 17, … 121, … 241); unlike the Video panel, the workflows do not snap other values, and the run fails at the Denoise node. The frame rate sets how much audio is generated, as well as the clip's duration.
* **Extending:** **Context Frames** defaults to 17 (about 0.7 s at 24 fps) — enough for motion, but only about one syllable of sound. For singing or other rhythmic material, raise it along the same grid (25, 33, 41, 49 …). **Number of Frames** *includes* the context, so the extension adds Number of Frames − Context Frames new frames; raise both together.
* **The Distilled LoRA** (the 8-step fast path for the Dev transformer) goes in the LoRA Collection workflow's list. Also set **Schedule** to `distilled`, **Steps** to 8 and **CFG** to 1 in the form — the `auto` schedule follows the checkpoint, not the LoRA.
The **Notes** panel of each lists the models to select in the form's **Model** section: an LTX-2.5 transformer for **Model**, the LTX-2.5 Components install for **Components**, and the Gemma-4 text encoder for **Text Encoder**.
:::
---
## Running a workflow
1. **Open the Workflows tab** and load one of the workflows from the library.
2. **Read the Notes** (note icon) and select the models it lists — the main model, VAE, and text encoder, plus the Lightning LoRA pair on the A14B workflows, the H3 loader's three selections plus the Turbo LoRA on the MiniMax workflows, or the LTX-2 loader's three selections on the LTX-2 workflows.
3. **Set your inputs:**
- *Text to Video* — type a **positive prompt** describing the scene and motion.
- *Image to Video / First (or Last) Frame to Video* — drop a starting (or ending) **image** into the image field, and optionally add a prompt.
- *Extend Video* — provide a **starting video** and trim it with the **Start Frame** and **End Frame** sliders (they appear once the video is loaded); the continuation picks up from the end of the trimmed range.
4. **(Optional) adjust resolution.** The defaults are chosen to balance speed and memory. If you hit an out-of-memory error, lower the resolution preset — see the [VRAM tips](#which-one-fits-my-gpu) below.
5. **Press Invoke.** The finished MP4 appears in the gallery and plays inline in the viewer — with its audio track, on H3.
:::tip
Leave the **frame count** and **steps** at their defaults to start. They're already tuned for each workflow (for example, the Lightning workflows use only 4 steps and the H3 workflows 6, which is what makes them fast).
:::
---
## Which one fits my GPU?
A rough guide — exact numbers depend on resolution and your system. See [Video Generation → OOM errors](/users-guide/video-generation/introduction/#oom-errors) for the full picture.
| Your GPU | Best choice |
|---|---|
| 24 GB and up | Any Wan workflow at 720p; the MiniMax H3 workflows (with `enable_partial_loading` and generous system RAM). |
| 16–24 GB | A14B Lightning workflows at 480p; TI2V-5B at up to 720p. |
| 12–16 GB | The **TI2V-5B (Low Quality)** workflows. |
If you get an **out-of-memory** error, lower the resolution first (it helps far more than reducing the frame count), and switch to a TI2V-5B workflow if you're still over budget. On H3, switch the resolution to **768 lowres** for drafts.
---
## Going further
These workflows are starting points. Once you're comfortable, the [Video Generation guide](/users-guide/video-generation/introduction/) covers the **Video panel** (the no-graph way to run all of this), the model variants, how image conditioning works, extending clips into longer videos, recommended parameters, and troubleshooting.
Read the Video Generation guide