Exercise 2: What Denoising Strength Actually Trades Off — Possible Solution
====================================================================

WHAT PURE-NOISE GENERATION LOOKS LIKE, PER imgai1-2
------------------------------
Per imgai1-2, ordinary generation "start[s] from pure random noise,
run[s] the trained network's denoising step repeatedly, [and a]
coherent image emerge[s] from static" — the starting point carries no
information from any real image at all; everything in the final result
comes from the denoising process being steered by the text embedding
alone.

WHAT img2img CHANGES ABOUT THE STARTING POINT
------------------------------
Per this chapter, img2img instead "starts from a real, existing image
with only a partial amount of noise added, then runs the same trained
denoising steps from there." Unlike pure noise, a partially-noised real
image still retains real structural information from the original —
edges, overall composition, rough color layout — that survives
however much noise was actually added.

WHY LOW DENOISING STRENGTH KEEPS THE ORIGINAL DOMINANT
------------------------------
Per this chapter, "a low value keeps the result very close to the
original image (only lightly re-touched by the prompt)." A low
denoising strength means only a small amount of noise was added to the
starting image, so the network's own denoising steps have very little
actual "static" to remove or rebuild — most of the original image's
own real structure survives untouched, and the text prompt can only
make small, localized adjustments to what's already there.

WHY HIGH DENOISING STRENGTH LETS THE PROMPT DOMINATE
------------------------------
Per this chapter, "a high value approaches full, pure-noise generation,
where the original image's influence becomes minimal." A high
denoising strength means a large amount of noise was added, destroying
most of the original image's own real structural information before
denoising ever starts. With little real signal left from the original
to anchor the process, the trained denoising steps end up rebuilding
the image almost entirely according to the text prompt's own steering —
functionally converging toward ordinary from-pure-noise generation
(imgai1-2), just with a faint residual influence from wherever the
partially-destroyed original still shows through.

WHY THIS WORKS AS AN ANSWER
------------------------------
It connects denoising strength directly to how much of the starting
image's real structural information survives the initial noise
addition, and explains both extremes (low = mostly original, high =
mostly prompt-driven) as a direct, mechanical consequence of how much
of that real structure remains for the denoising process to build from.