Strict mirroring版本IL V2 (ID: 1276456)

This is early experimental LoRa for researching borders of training the concept of strict mirror.

I think it is possible because SDXL can draw mirror floors.

Status

IL V2: this LoRa version can increase probability (+24% for 8-20 steps) of getting strict reflections in a new generation. Increasing steps is refining the results. Increasing LoRa's weight can refine the results (60% success with weight=2.0).

IL V1: this LoRa version can slightly increase probability (+10% for 4 dmd2-steps, +12-16% for 8 dmd2-steps) to find strict reflections in a new generation. Increasing steps is refining the results.

I go to the next step of the experiment.

Changelog

IL V2

• Added negative tokens:

  • x_out_of_mirror (protects against a reflection outside mirror's bounds)

  • x_differect_reflection (supress a wrong symmetry)

  • different reflection (same the x_out_of_mirror but danbooru tag)

• Added ai-generated images with negative tokens to dataset.

• 60 epochs instead of 10 epochs of version 1

• Trained with cleaned dictionary: x_mirror_reflection, x_reflection_face, x_reflection_facing_aside, x_reflection_facing_away, reflection, mirror, hand on mirror, full-length mirror, broken mirror, reflection focus, looking at mirror.

Usage

Connect LoRa in your workflow and then activate by word x_mirror_reflection in your prompt. Next choose one of these controls for the position of the character's face relative to the mirror:

• x_reflection_facing_away

• x_reflection_facing_aside

• x_reflection_face

Now you can use negative prompt (+8% to find good mirroring in IL V2): x_out_of_mirror, x_differect_reflection, different reflection

描述:

VERSION 2

CHANGE LOG

• Added negative tokens:

  • x_out_of_mirror (protects against a reflection outside mirror's bounds)

  • x_differect_reflection (supress a wrong symmetry)

  • different reflection (same the x_out_of_mirror but danbooru tag)

• Added ai-generated images with negative tokens to dataset.

• 60 epochs instead of 10 epochs of version 1

• Trained with cleaned dictionary: x_mirror_reflection, x_reflection_face, x_reflection_facing_aside, x_reflection_facing_away, reflection, mirror, hand on mirror, full-length mirror, broken mirror, reflection focus, looking at mirror.

--------------------------

TESTS

--------------------------

holy mix illustrous, spo, dmd2, 8 steps, 1 cfg, with mirror lora, 25 runs

--------------------------

pos. prompt:

--------------------------

masterpiece, best quality, amazing quality, very aesthetic, absurdres, newest,

1girl, solo,

(((x_mirror_reflection))), reflection, mirror,

{||hand on mirror},

{|||full-length mirror},

{||broken mirror}

{||reflection focus},

{x_reflection_face, looking at mirror|x_reflection_facing_aside, {looking at viewer|looking at another|looking back, looking to the side, facing away}|x_reflection_facing_away, {looking at viewer|looking at another|looking back, looking to the side, facing away}}

--------------------------

32% - strict reflection

16% - strict reflection but outside mirror borders

48% - wrong symmetry

4% - trash (prompt misunderstanding)

--------------------------

--------------------------

holy mix illustrous, spo, dmd2, 8 steps, 1 cfg, no mirror lora, 25 runs,

--------------------------

pos. prompt: same

--------------------------

8% - strict reflection

4% - strict reflection but outside mirror borders

56% - wrong symmetry

32% - trash (prompt misunderstanding)

--------------------------

--------------------------

holy mix illustrous, spo, 20 steps, 6 cfg, with mirror lora, 25 runs,

--------------------------

pos. prompt: same

neg. prompt: x_out_of_mirror, x_differect_reflection, different reflection

--------------------------

40% - strict reflection

16% - strict reflection but outside mirror borders

40% - wrong symmetry

4% - trash (prompt misunderstanding)

--------------------------

--------------------------

holy mix illustrous, spo, 20 steps, 6 cfg, without mirror lora, 25 runs,

--------------------------

pos. prompt: same

neg. prompt: same

--------------------------

16% - strict reflection

0% - strict reflection but outside mirror borders

16% - wrong symmetry

68% - trash (prompt misunderstanding)

--------------------------

Hypothesis: lora weight x2 can refine results.

-----------------------------

Illustrous, spo, dmd2, 8 steps, 1.5 cfg, with mirror lora, 25 runs

--------------------------

pos. prompt: same

neg. prompt: same

--------------------------

32% - strict reflection

20% - strict reflection but outside mirror borders

36% - wrong symmetry

12% - trash (prompt misunderstanding)

--------------------------

-----------------------------

Illustrous, spo, dmd2, 8 steps, 1.5 cfg, without mirror lora, 25 runs

--------------------------

pos. prompt: same

neg. prompt: same

--------------------------

12% - strict reflection

0% - strict reflection but outside mirror borders

32% - wrong symmetry

56% - trash (prompt misunderstanding)

--------------------------

-----------------------------

Illustrous, spo, dmd2, 8 steps, 1.5 cfg, with mirror lora x2 strength, 25 runs

--------------------------

pos. prompt: same

neg. prompt: same

--------------------------

60% - strict reflection

2% - strict reflection but outside mirror borders

36% - wrong symmetry

0% - trash (prompt misunderstanding)

--------------------------

-----------------------------

Illustrous, spo, dmd2, 8 steps, 1.5 cfg, with mirror lora x3 strength, 50 runs

--------------------------

pos. prompt: same

neg. prompt: same

--------------------------

36% - strict reflection

0% - strict reflection but outside mirror borders

62% - wrong symmetry

2% - trash (prompt misunderstanding)

and dramatically low quality of all results in this batch.

--------------------------

SUMMARY

• This version increases probability of getting strict mirroring by 24%.

• Negative prompt increases probability of getting strict mirroring by 8%.

• Try to increase LoRa weight.

• High LoRa weight (>2) spoils results.

训练词语: x_mirror_reflection,x_reflection_face,x_reflection_facing_aside,x_reflection_facing_away