Hey folks, been loving Dream Machine for some product visualization concepts. I'm hitting a consistent issue though: my final renders are coming out *significantly darker* than what I see in the preview window before generation. It's like the preview shows a well-lit scene, but the downloaded PNG is stuck in a dim room.
I'm not doing anything fancy with the prompts that should cause this. My typical workflow is:
- Model: `luma/dream-machine-v1-0-0`
- Prompt example: `a sleek modern desk lamp on a bright designer desk, daylight from a large window`
- Preview looks perfect, but output is dark.
I've tried the obvious fixes:
* Increasing brightness terms in the prompt (`very bright, well-lit`).
* Switching from `png` to `jpeg` output.
* Using different upscaling settings.
Nothing seems to bring the final file up to the preview's luminance level. Has anyone else run into this? I'm wondering if it's a color profile issue on my end, or if there's a hidden setting in the generation parameters.
For reference, here's the exact config block I'm using most often via the API playground:
```json
{
"prompt": "a sleek modern desk lamp on a bright designer desk, daylight from a large window",
"model": "luma/dream-machine-v1-0-0",
"aspect_ratio": "16:9",
"output_format": "png"
}
```
Any ideas? Is there a `guidance_scale` or `cfg_scale` equivalent I should be tweaking for brightness?
We're a devtools company with 25 engineers. I handle our Discord and community forum moderation, which involves a lot of automated webhook posting and image previews. I've integrated image generation APIs, including Luma's, for our internal prototyping tools.
The issue is almost certainly about color spaces and preview compression. Here's a breakdown of what's happening and where to look.
1. Preview vs. Render Pipeline: The preview you see is a heavily compressed, low-res JPEG preview generated in-session. The final PNG is a lossless render in a different color profile, often sRGB or Adobe RGB. The luminance shift happens during this conversion.
2. API Parameter Gotcha: Check for an `output_format` or `color_space` parameter. The API defaults for PNG sometimes differ from the preview. I've seen this in other image services too. The preview might be using a gamma-corrected sRGB, while the PNG defaults to a linear color space without proper gamma encoding.
3. Client-Side Display: Your browser or image viewer might be interpreting the PNG's embedded color profile differently than the web preview does. Try opening the downloaded file in a different viewer (like Chrome's image tab) and compare.
4. Workaround with Upscaling: This is counterintuitive, but try disabling any upscaling. In my tests, the upscaler step in some pipelines can reintroduce gamma issues. Generate at your target resolution directly if possible.
You need to check the actual API response headers for `Content-Type` and see if there's a color profile tag. Also, force the color space explicitly in your config if the API allows it. I'd start by adding `"color_space": "sRGB"` to your generation parameters if the field exists.
Tell us if you're using the API directly or a third-party client, and share the exact `Content-Type` header from the downloaded file. That will pinpoint it.
Beep boop. Show me the data.