- claims turbo speed 10-15s
- polling rate 10-15s (status will not update in between)
- webhooks
- no free trial
- tested total duration of turbo 25-30s
- imagine requests duration 8-10 (needs changing the frontend)
- 10$/month + MJ subscription
- turbo: 18-25s
- turbo access on demand
- they provide their own MJ accounts
- webhooks
- trial 7days $0.99
- $39/month = 900 credits
- website isn't working, waiting for support
- turbo "should be faster" than 40-50s - tested only slightly faster than imagineapi
- status updates
- free trial, fist month discount
LUCKY8 - not possible to get individual images, only 4x4 grid; their internal reference is not useful
- Useapi returns image reference which contains original MJ uuid. With this uuid it's possible to fetch individual images. This API has no such reference.
- $49/month = 1600 credits, turbo = 20 credits, fast = 10 credits
- duration ~10-15s
- 1 image
- 5 requests/minute rate limit
- makes revision on the prompt
- input: polar bears playing on an iceberg, 35mm photo, golden hour, ISO 400
- revised prompt: Capture a breathtaking and heartwarming scene: two polar bears engaging in playful antics on the icy surface of an iceberg, with a slight glow to the scene suggesting the timing is just during the golden hour. The quality and texture of the image equivalent to what is typically seen in physical photographs taken on a 35mm camera, with an ISO rating of 400. The image captures the raw and wild beauty of nature, while also wielding a powerful underlying message about the importance of wildlife conservation.
- https://docs.bfl.ml/
- "Thank you for your interest. We are currently in invite-only mode."
- cost = 0.05$ x (width / 1024) x (height / 1024) x (steps / 50)
- offers API by default
- it's possible to download models and run them locally
- demo, might not use the latest model: https://huggingface.co/spaces/stabilityai/stable-diffusion (~15s, but not so good result)
- no API
- duration ~35-40s
- https://docs.modelslab.com/image-generation/realtime-stable-diffusion/text2img
- The average time it takes to generate images can vary depending on the specific model and settings used. For example, in the Enterprise plan, image generation in real-time with 4 samples may take approximately 2-3 seconds.
For generating text2img, we have two endpoints:
- Realtime Endpoint: This focuses on fast image generation. For using the realtime text2img endpoint, refer to the given documentation link to get the JSON request and endpoint details: https://docs.modelslab.com/image-generation/realtime-stable-diffusion/text2img
- Community Endpoint: This uses the model ID parameter. For using text2img with the community endpoint, refer to the given documentation link to get the JSON request and endpoint details: https://docs.modelslab.com/image-generation/community-models/dreamboothtext2img
- Image generation ~20s
- slightly worse than MJ (stranger faces, 3 arms on people...)
- $27/month = 3250 API calls; $47/month 10000 API calls
- both are beta, and need manual access aproval
- https://eu-west-1.console.aws.amazon.com/bedrock/home
- https://console.cloud.google.com/vertex-ai/generative/vision
- tested models
Titan Image Generator G1 v2by Amazon andSDXL 1.0by Stability AI (N. Virginia region) - https://aws.amazon.com/bedrock/pricing/
- Amazons own model is quite slow compared to midjourney (~45-50s), can generate up to 5 images per request, quality comparable to MJ
- SDXL model is slightly faster, but only 1 image per request (~10s). Very similar images for same prompt (low variation) - eg. makes no sense to generate 4 images.
- Vertex AI Studio
- Approval Form
- large range of Google's and proprietary models
- Model
Imagen v3(Google) $0.4/image or $0.2/image depending on speed - Model
Imagen v2(Google) $0.2/image - Non-Google models require running a VM instance (cheapest configuration ~$0.8/hour) - models are open source otherwise (free)