fal-realtime

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

fal-realtime

Real-time and streaming image generation using fal.ai's fastest models. Results in under 1 second — ideal for interactive applications, live previews, and rapid iteration.

使用fal.ai最快模型实现的实时流式图像生成。不到1秒即可出结果，非常适合交互式应用、实时预览和快速迭代。

Scripts

脚本

Script	Purpose
`realtime.sh`	Generate images in real-time (sub-second)

脚本	用途
`realtime.sh`	实时生成图像（亚秒级）

Usage

使用方法

Quick Real-Time Generation

快速实时生成

bash

./scripts/realtime.sh --prompt "A cute cat wearing a top hat"

bash

./scripts/realtime.sh --prompt "A cute cat wearing a top hat"

With LoRA

配合LoRA使用

bash

./scripts/realtime.sh --prompt "A portrait in sks style" --lora-url "https://example.com/lora.safetensors"

bash

./scripts/realtime.sh --prompt "A portrait in sks style" --lora-url "https://example.com/lora.safetensors"

Multiple Rapid Iterations

多次快速迭代

bash

./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 1"
./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 2, more vibrant"
./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 3, darker tones"

bash

./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 1"
./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 2, more vibrant"
./scripts/realtime.sh --prompt "Abstract art with flowing colors, version 3, darker tones"

Arguments

参数说明

Argument	Description	Required
`--prompt` / `-p`	Image description	Yes
`--model` / `-m`	Model endpoint	No (default: fal-ai/flux-2/klein/realtime)
`--size`	square, landscape, portrait	No (default: square)
`--seed`	Random seed for reproducibility	No
`--lora-url`	URL to LoRA weights	No
`--lora-scale`	LoRA influence (0.0-1.0, default: 1.0)	No
`--num-images`	Number of images (default: 1)	No

参数	描述	是否必填
`--prompt` / `-p`	图像描述	是
`--model` / `-m`	模型端点	否（默认值：fal-ai/flux-2/klein/realtime）
`--size`	尺寸，可选square（正方形）、landscape（横版）、portrait（竖版）	否（默认值：square）
`--seed`	用于复现结果的随机种子	否
`--lora-url`	LoRA权重的URL	否
`--lora-scale`	LoRA影响权重（取值范围0.0-1.0，默认值：1.0）	否
`--num-images`	生成图像数量（默认值：1）	否

Finding Models

查找模型

To discover the best and latest real-time generation models, use the search API:

bash

undefined

要发现最佳的最新实时生成模型，可使用搜索API：

bash

undefined

Search for real-time / fast image generation models

搜索实时/快速图像生成模型

bash /mnt/skills/user/fal-generate/scripts/search-models.sh --query "realtime" bash /mnt/skills/user/fal-generate/scripts/search-models.sh --query "fast image" bash /mnt/skills/user/fal-generate/scripts/search-models.sh --category "text-to-image"


Or use the `search_models` MCP tool with keywords like "realtime", "fast", "schnell", "turbo".


或者使用`search_models` MCP工具，搭配“realtime”、“fast”、“schnell”、“turbo”这类关键词搜索。

When to Use Real-Time vs Standard Generation

实时生成 vs 标准生成适用场景

Use Case	Use Real-Time	Use Standard
Rapid prototyping / iteration	Yes
Interactive apps / live preview	Yes
Final high-quality output		Yes (FLUX Dev, Nano Banana Pro)
Professional / commercial work		Yes (FLUX Pro Ultra)
Exploring prompt ideas quickly	Yes
Batch generation with variations	Yes

适用场景	选择实时生成	选择标准生成
快速原型设计/迭代	是
交互式应用/实时预览	是
最终高质量输出		是（FLUX Dev、Nano Banana Pro）
专业/商业作品		是（FLUX Pro Ultra）
快速探索提示词思路	是
带变体的批量生成	是

Output Format

输出格式

json

{
  "images": [
    {
      "url": "https://fal.media/files/...",
      "content_type": "image/jpeg",
      "width": 1024,
      "height": 1024
    }
  ],
  "seed": 12345,
  "has_nsfw_concepts": [false]
}

json

{
  "images": [
    {
      "url": "https://fal.media/files/...",
      "content_type": "image/jpeg",
      "width": 1024,
      "height": 1024
    }
  ],
  "seed": 12345,
  "has_nsfw_concepts": [false]
}

Tips

使用提示

Real-time models use fewer inference steps — prompts should be clear and concise
Use
```
--seed
```
to lock in a good composition, then iterate on the prompt
LoRA support works with FLUX Klein Realtime — great for style-locked fast iteration
These models use the synchronous API (no queue), so results are instant

实时模型使用更少的推理步骤，提示词应清晰简洁
可使用
```
--seed
```
锁定满意的构图，再针对提示词进行迭代
FLUX Klein Realtime支持LoRA，非常适合固定风格的快速迭代
这些模型使用同步API（无需排队），因此可即时返回结果