The trigger word is 'gwm txxx'.

Supports generating high-resolution images now!

• 1664×936, 16:9

• 1280×960, 4:3

• 1024×1024, 1:1

• or other you like

Better scenery and vehicle photography composition.

Better vehicle details.

The problem of word pollution is alleviated.

Recommended params

• Sampling method: DPM++ 2M SDE Karras

• CFG Scale: 3~5

• Checkpoint: Juggernaut XL - V 7 + RunDiffusion | Stable Diffusion Checkpoint | Civitai ( Highly Recommended! )

• Prompt:

  • {your description}, ultra detail, Eye-level product film photography, depth of field, shot on 120mm, shot on Hasselblad, Cross Processing Color grading

  • {your description}, ultra detail, beautiful, hd, 8k, (photographic:1.5), magnificent, grandiose, macos wallpaper

• Negative prompt:

  • collapse, scatter, out of frame, low-res, text, error, cropped, worst quality, low quality, jpeg artifacts, duplicate, wrong proportions, mutation, deformed, blurry, bad anatomy, bad proportions, username, watermark, signature, (low quality, worst quality:1.3)

Note: it may appear to collapse more frequently when generating an image resolution of 1664 × 936(or even bigger). We guess the resolution gap between training data and the test leads to this phenomenon(the largest size of our training input is 1360 × 768).

Any suggestion is appreciated！