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

# Overview

> Compare and choose the right autoregressive video diffusion pipeline for your use case

# Pipelines Supported in Scope

Scope supports six video pipelines. Five are autoregressive video diffusion pipelines built on Wan2.1 base models using the [Self-Forcing](https://self-forcing.github.io/) training methodology (except RewardForcing, which uses Rewarded Distribution Matching). The sixth, **LTX 2.3**, is a joint audio-video DiT model from Lightricks that ships as an installable plugin rather than a built-in pipeline.

Each pipeline has its own strengths. Some optimize for speed, others for quality, long-context consistency, or synchronized audio. The comparison tables below summarize the key differences to help you choose.

***

## Quick Comparison

Four pipelines use the smaller Wan2.1 1.3B model, Krea Realtime uses the larger Wan2.1 14B model for higher quality output, and LTX 2.3 uses a 22B DiT from Lightricks.

| Pipeline                                                            | Base Model               | Creator                              | Estimated VRAM\* |
| :------------------------------------------------------------------ | :----------------------- | :----------------------------------- | ---------------: |
| [StreamDiffusion V2](/scope/reference/pipelines/streamdiffusion-v2) | Wan2.1 1.3B              | StreamDiffusion team                 |           \~20GB |
| [LongLive](/scope/reference/pipelines/longlive)                     | Wan2.1 1.3B              | Nvidia, MIT, HKUST, HKU, THU         |           \~20GB |
| [Krea Realtime](/scope/reference/pipelines/krea-realtime)           | Wan2.1 14B               | Krea                                 |           \~32GB |
| [RewardForcing](/scope/reference/pipelines/reward-forcing)          | Wan2.1 1.3B              | ZJU, Ant Group, SIAS-ZJU, HUST, SJTU |           \~20GB |
| [MemFlow](/scope/reference/pipelines/memflow)                       | Wan2.1 1.3B              | Kling                                |           \~20GB |
| [LTX 2.3](/scope/reference/pipelines/ltx-2-3)                       | Lightricks DiT 22B (FP8) | Lightricks                           |           \~22GB |

\*Estimated runtime VRAM usage. A 24GB GPU (e.g. RTX 4090) is the minimum commercially available card for 1.3B pipelines and LTX 2.3. See [System Requirements](/scope/reference/system-requirements) for minimum hardware specs.

***

## Feature Support

All five Wan2.1-based pipelines support Text-to-Video (T2V), Video-to-Video (V2V), LoRA adapters, and VACE conditioning. MemFlow adds a memory bank for long-context consistency. LTX 2.3 is the only pipeline that generates synchronized audio alongside video, but it does not support VACE.

| Feature              | StreamDiffusion V2 |  LongLive  | Krea Realtime | RewardForcing |   MemFlow  |    LTX 2.3    |
| :------------------- | :----------------: | :--------: | :-----------: | :-----------: | :--------: | :-----------: |
| Text-to-Video (T2V)  |         Yes        |     Yes    |      Yes      |      Yes      |     Yes    |      Yes      |
| Image-to-Video (I2V) |         No         |     No     |       No      |       No      |     No     |      Yes      |
| Video-to-Video (V2V) |         Yes        |     Yes    |   Limited\*   |      Yes      |     Yes    |  Via IC-LoRA  |
| Synchronized Audio   |         No         |     No     |       No      |       No      |     No     |      Yes      |
| LoRA Support         |     1.3B LoRAs     | 1.3B LoRAs |   14B LoRAs   |   1.3B LoRAs  | 1.3B LoRAs | LTX 2.3 LoRAs |
| VACE Support         |         Yes        |     Yes    |      Yes      |      Yes      |     Yes    |       No      |
| Memory Bank          |         No         |     No     |       No      |       No      |     Yes    |       No      |

<Warning>
  \*Krea Realtime's regular V2V mode (latent initialization) has known quality issues. Use **VACE V2V** (visual conditioning with input video) for better results.
</Warning>

***

## Pipeline Details

<CardGroup cols={3}>
  <Card title="StreamDiffusion V2" href="/scope/reference/pipelines/streamdiffusion-v2">
    Real-time streaming from the original StreamDiffusion creators
  </Card>

  <Card title="LongLive" href="/scope/reference/pipelines/longlive">
    Smooth prompt transitions and extended generation from Nvidia
  </Card>

  <Card title="Krea Realtime" href="/scope/reference/pipelines/krea-realtime">
    14B model for highest quality generation
  </Card>
</CardGroup>

<CardGroup cols={3}>
  <Card title="RewardForcing" href="/scope/reference/pipelines/reward-forcing">
    Reward-matched training for improved output quality
  </Card>

  <Card title="MemFlow" href="/scope/reference/pipelines/memflow">
    Memory bank for long-context consistency
  </Card>

  <Card title="LTX 2.3" href="/scope/reference/pipelines/ltx-2-3">
    Joint audio-video generation with ID-LoRA and IC-LoRA control (plugin)
  </Card>
</CardGroup>

***

## Shared Parameters

The Wan2.1-based pipelines share these common parameters. LTX 2.3 uses a different set of constraints (32-pixel resolution multiples and 8×K+1 frame counts), documented on its [reference page](/scope/reference/pipelines/ltx-2-3#output-constraints).

### Resolution

Generation is faster at smaller resolutions, resulting in smoother video. The visual quality is best at **832x480**, which is the training resolution for most pipelines. You may need a more powerful GPU to maintain high FPS at this resolution.

### Seed

The seed parameter enables reproducible generations. If you find a seed value that produces good results with a specific prompt sequence, save it to reproduce that generation later.

***

## Prompting

These techniques apply to all pipelines and significantly improve output quality.

### Subject and Background Anchors

Include a clear **subject** (who/what) and **background/setting** (where) in each prompt. For scene continuity, reference the same subject and/or setting across prompts.

```
"A 3D animated scene. A panda walks along a path towards the camera in a park on a spring day."
```

```
"A 3D animated scene. A panda halts along a path in a park on a spring day."
```

### Cinematic Long Takes

The models work better with **long cinematic takes** rather than rapid shot-by-shot transitions. Avoid quick cutscenes, rapid scene changes, and jump cuts. Instead, let scenes flow naturally with gradual transitions.

### Long, Detailed Prompts

All pipelines perform better with **detailed prompts**. A helpful technique is to expand a base prompt using an LLM (ChatGPT, Claude, etc.).

**Base prompt:**

```
"A cartoon dog jumping and then running."
```

**Expanded prompt:**

```
"A cartoon dog with big expressive eyes and floppy ears suddenly leaps into the frame, tail wagging, and then sprints joyfully toward the camera. Its oversized paws pound playfully on the ground, tongue hanging out in excitement. The animation style is colorful, smooth, and bouncy, with exaggerated motion to emphasize energy and fun. The background blurs slightly with speed lines, giving a lively, comic-style effect as if the dog is about to jump right into the viewer."
```

***

## See Also

<CardGroup cols={3}>
  <Card title="VAE Types" icon="cube" href="/scope/reference/vae">
    Configure VAE for quality/speed tradeoffs
  </Card>

  <Card title="System Requirements" icon="server" href="/scope/reference/system-requirements">
    Hardware requirements for each pipeline
  </Card>

  <Card title="Pipeline Architecture" icon="sitemap" href="/scope/reference/architecture/pipelines">
    Technical details for node developers
  </Card>
</CardGroup>