Dreamlook
Lightning-fast Stable Diffusion finetuning platform for training models and generating images at scale.
Last verified:
What is Dreamlook?
Dreamlook.ai is an advanced AI platform designed to make Stable Diffusion model training and image generation faster, easier, and scalable. It supports full-model training (not just LoRA) for SD 1.5 and SDXL, delivering high performance and exceptional results in just minutes. The platform trains models 2.5x faster than competitors, capable of training 1,500 SDXL steps in only 10 minutes with no quality compromise.
Key features include unbeatable training performance, creating full-res SDXL images in just 4 seconds, LoRA file extraction to reduce downloaded file size to 75 MB, and a robust scalable finetuning API for powering applications without worrying about GPU quotas or spinning up instances. The platform finetunes both text encoder and UNet, giving equivalent results to diffusers implementation. Users can train on anything including objects, styles, or specific subjects using instance prompts like
Dreamlook pricing
Pricing model: Paid
Finetuning Stable Diffusion 1.5 starts at $0.75/run, and SDXL at $2.25/run. Generating 1024x1024 images costs from $0.01/image for Stable Diffusion 1.5, and $0.02/image for SDXL. Packages include: $15 (up to 15 SD1.5 training runs, 5 SDXL training jobs, 1.2k generated images), $60 (up to 70 SD1.5 runs, 23 SDXL jobs, 5.6k images), $100 most popular (up to 120 SD1.5 runs, 40 SDXL jobs, 9.6k images), $400 (up to 500 SD1.5 runs, 166 SDXL runs, 40k images), $750 (up to 1,000 SD1.5 runs, 333 SDXL jobs, 80k images). New users get 50 free tokens on signup with no credit card required.
Dreamlook pros
- Trains 2.5x faster than competitors
- 1,500 SDXL steps in just 10 minutes
- Full-res SDXL images created in 4 seconds
- Full model finetuning, not just LoRA
- LoRA file extraction reduces size to 75 MB
- Robust scalable finetuning API available
- No GPU quotas or instance management needed
- Supports both SD1.5 and SDXL models
- Train on objects, styles, or specific subjects
- Data automatically deleted after 48 hours for privacy
- Trained models belong to you with no lock-in
- Download and use models anywhere
- Easy-to-use interface for non-developers
- Competitive pricing starting at $0.75/run
- 50 free tokens on signup to start
- Supports ControlNet and SDXL features
- No credit card required to start training
Dreamlook cons
- Models only available for 48h by default after training
- No native LoRA finetuning offered
- No free trial package available
- Cannot train or generate images involving nudity
- Extracted LoRA results not as good as full model training
- Limited free tier options
- Training images showing people must be adults with permission
- No PayPal support yet (only Stripe)
- Wait times may be longer than anticipated during beta
Frequently asked questions about Dreamlook
Can I train a model on objects or styles?
Yes, you can train on anything you want! Use the instance prompt for this. For example, use 'ukj style' for a style or 'photo of ukj object' for an object.
What happens to the images I upload?
They're deleted after 48 hours, along with the trained models. For privacy and security reasons, data is quickly deleted from servers including both uploaded images and trained models. You can extend storage duration by subscribing to one of the paid plans.
How are the models trained?
Dreamlook.ai is finetuning both text encoder and UNet. The method gives equivalent results to the diffusers implementation. At this time, regularization images (also known as prior preservation) are not used.
Can I train or generate images involving nudity?
No, per the Terms of Use, you are not allowed to train from or generate images involving nudity. Training images showing people must show adults who have given their explicit permission.
Why use dreamlook.ai?
Privacy: Data kept maximum 48h by default then deleted. No lock-in: Download trained custom models and use them for whatever you want. Speed: Fastest finetuning service with equivalent results to other platforms using state-of-the-art infrastructure. Ease-of-use: No developer skills required but developers will love it too. Competitive pricing with first model free then pay as you go.
How do I reach out?
You can join their Discord server or reach out to [email protected] for customer service.
What base models are available?
Available base models include Stable Diffusion 1.5, Realistic Vision, Analog Diffusion, Dreamlike Photoreal v2, and RPG-v4.
Do extracted LoRAs give good results?
Extracted LoRAs give better results than native LoRAs. They require more resources as you need to train the full model first, but Dreamlook.ai does this very efficiently with models trained in minutes. Extracted LoRAs have rank 64 resulting in a 75 MB file.
Is there extra cost to extract LoRA files?
No, there is no extra cost to extract a LoRA file when training a model on Dreamlook.ai and it is enabled by default. At the end of your run you should see both full checkpoint and LoRA appear. For API users, set extract_lora to 'original' (disabled by default).
How long does training take?
Dreamlook.ai trains 1,200 steps in just 3 minutes (training text encoder + unet). The platform can train 1,500 SDXL steps in only 10 minutes. Typically takes about 15 minutes to generate an image after images have been uploaded for avatar creation.