I am bummed it didn’t work, I lost the res

More about I am bummed it didn’t work, I lost the res

How the Images Match the Process

This image grid shows a two-pass Deep Style workflow. Each letter (a, b, c) represents a different role in the process.

⸻

Base Images (a)

1a, 2a, 3a, 4a, 5a, 6a, 7a, 8a, 9a, 10a

These are the original base images.
Nothing has been done to them yet.
Faces are ideal because the AI strongly recognizes eyes, mouths, symmetry, and head shape.

⸻

Filter / Texture Sources (b)

1b, 2b, 3b, 4b, 5b, 6b, 7b, 8b

These are not results.
They are textures or generated noise patterns used as edge-builders:
• surface grain
• structured contrast
• repeating patterns
• directional noise

Their job is to create edges, not to look good.

⸻

First Deep Style Pass (c)

1c, 2c, 3c, 4c, 5c, 6c, 7c, 8c

Here the base image (a) is processed using the filter (b).
The AI begins detecting edges, but:
• faces may distort
• structure is unstable
• detail is incomplete

This output is not final.
It is a filter layer, not an artwork.

⸻

Manual Layer Recombination (outside Deep Dream)

(This step is not a single image—it happens between rows.)

Process:
• Open the original base image
• Place the Deep Style result on a layer above it
• Adjust opacity and blend mode
• Mask or erase areas (especially faces)

This step lets you control where edges exist.

⸻

Second Deep Style Pass

9b, 10b, 10c

The recombined image is run through Deep Style again.
Now the AI:
• sees clearer edges
• locks onto facial structure
• adds detail instead of chaos

This is where intelligence appears.

⸻

Detail Proof (eyes)

11a, 11b

Close-ups show:
• stable eye shape
• controlled texture scale
• edges aligned with anatomy

This confirms successful edge guidance.

⸻

Key
• a = base image
• b = filter / edge generator
• c = Deep Style output

⸻

Summary

This is not decoration.
You are constructing edges for the AI, recombining them manually, and feeding them back so the model can interpret the image more accurately on a second pass.