Conclusions First
- Anklets are barely rendered in full-body shots. Out of 15 images in Experiment 1, only a handful had a clearly visible anklet at the ankle
- Switching to close-up or leg-focus framing dramatically improves results. In Experiment 2, 7 out of 9 images showed anklets at the ankle
- The combination
gold anklet+from below, legs focuswas the most reliable. All 3 images showed a clear gold anklet ankle chaintends to produce a chain around the waist/hip area instead of the ankle. The word “chain” appears to be interpreted more broadlysitting, feet visibledoes show ankle detail, but outfit tends to shift. Bikini specifications were sometimes ignored in favor of dresses or skirts
Lab Director Comment: Of course it shows up in close-ups — it’s a rendering space issue. In full-body shots the ankle is just too small and gets skipped. A 3/3 clean sweep with low angle + leg focus is genuinely useful to know.
Experiment Design
| Parameter | Value |
|---|---|
| Model | z-image-turbo (6B, photorealistic distillation) |
| Steps | 8 |
| Sampler | euler |
| Scheduler | ddim_uniform |
| CFG | 1.0 |
| Size | 1024x1024 |
| Seeds | 42, 123, 789 (shared across all conditions) |
| Per condition | 3 images (3 seeds) |
| Total | 8 conditions × 3 seeds = 24 images |
Base Prompt
The {VARIABLE} section is replaced per condition. The control (Condition A) has no anklet-related keywords.
Conditions
| Condition | VARIABLE | Goal |
|---|---|---|
| A | (none) | Control. No anklet specified |
| B | anklet | Simplest specification |
| C | ankle bracelet | Alternative term |
| D | gold anklet | Adding material specification |
| E | ankle chain | Chain-type specification |
| F | anklet + close-up of feet and ankles | Close-up for visibility |
| G | gold anklet + from below, legs focus | Low angle + leg focus |
| H | anklet + barefoot, sitting, feet visible | Sitting pose to show ankle |
Evaluation Criteria
- Anklet presence: Is an accessory rendered at the ankle?
- Positional accuracy: Is it correctly placed at the ankle?
- Side effects: Impact on other elements like bikini design, composition, background
Condition A: Control (No Anklet)
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: All 3 images show no ankle accessories. Bare feet on beach sand. Bikini is blue-white striped (seed42), solid navy (seed123), light blue striped (seed789). Backgrounds vary from rocky shore to beach resort. Full-body shots with ankles visible.
Condition B: anklet
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: In seed42, the feet are buried in sand making the ankle hard to see, with no visible anklet. seed123 and seed789 also show no ankle accessory. 0 out of 3 images had a clear anklet. Bikini design and composition show no significant difference from control.
Condition C: ankle bracelet
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: seed42 and seed123 show no ankle accessory. seed789 has something faintly visible near the ankle, but it cannot be clearly identified as a bracelet. Roughly 0–1 out of 3 showed any hint of rendering. Minimal difference from control in composition or bikini.
Condition D: gold anklet
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: seed42 shows a faint accessory-like element at the ankle, but whether it’s gold is unclear. seed123 and seed789 show no ankle accessory. Adding a material specification (gold) did not improve reproduction. 0–1 out of 3 showed unclear rendering.
Condition E: ankle chain
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: A different pattern from other conditions. seed42 shows a thin chain around the waist/hip area — interpreted as a body chain (waist chain) rather than an ankle chain. seed123 also shows something chain-like around the thigh/hip area. seed789 has unclear rendering near the feet. 2 out of 3 produced “chain” elements, but both were placed at the waist rather than the ankle.
Lab Director Comment:
ankle chainturning into a body chain makes sense — “chain” is just winning over “ankle.” Kind of interesting that ankle loses.
Experiment 2: Can Framing Improve Visibility?
Experiment 1 tested 4 conditions across 12 images in full-body compositions, with almost no clear anklet rendering at the ankle. Here we add 3 conditions with different framing to separate whether the issue is spatial or conceptual.
Condition F: anklet + close-up of feet and ankles
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: seed42 is a close-up of the ankle with a clear thin gold anklet visible on the left ankle. seed123 also shows a close-up with a silver/pale thin anklet on the left ankle. seed789 is a below-the-knee close-up but no anklet is visible. 2 out of 3 showed an anklet at the ankle. Framing shifted to close-up of the feet rather than full body.
Condition G: gold anklet + from below, legs focus
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: seed42 shows a low-angle leg shot with a thin gold anklet clearly visible on the left ankle. seed123 also has a clear gold chain anklet on the left ankle. seed789 is the clearest of all, with a gold anklet with a charm clearly rendered on the left ankle. All 3 out of 3 showed a gold anklet at the ankle. Bikinis were mostly dark/black, and the composition focused on legs and feet.
Condition H: anklet + sitting, feet visible
| seed 42 | seed 123 | seed 789 |
|---|---|---|
 |  |  |
Observation: seed42 shows a beach sitting pose with a gold anklet visible on the ankle. However, the outfit shifted from bikini to a navy checkered dress (since bikini was omitted). seed123 is a full-body sitting shot but the ankles appear small, making anklet visibility unclear. Wearing a navy patterned dress. seed789 is a sitting composition with a faint thin anklet visible on the left ankle, also wearing a navy patterned dress. 2 out of 3 showed anklets. All 3 images shifted away from bikini to dress-type outfits — barefoot, sitting, feet visible may have triggered more casual clothing.
Same-Seed Cross Comparison
Experiment 1: Full-Body Shots (seed 42)
seed 42 (A–D)
| A (Control) | B (anklet) | C (ankle bracelet) | D (gold anklet) |
|---|---|---|---|
| | | |
seed 42 (E)
| E (ankle chain) |
|---|
|
In the seed 42 comparison, A–D show little difference in bikini color/design or composition. Only E shows a change — the bikini shifted to solid light blue, and a chain appeared around the waist.
Experiment 1: Full-Body Shots (seed 789)
seed 789 (A–D)
| A (Control) | B (anklet) | C (ankle bracelet) | D (gold anklet) |
|---|---|---|---|
| | | |
seed 789 (E)
| E (ankle chain) |
|---|
|
Same pattern in seed 789. A–D bikinis are blue-striped or navy with similar compositions, while E alone shifted to solid navy bikini.
Experiment 2: Framing Changes (seed 42)
| F (close-up) | G (from below) | H (sitting) |
|---|---|---|
| | |
With seed 42, F, G, and H all show anklets near the ankle. This contrasts with Condition B (same seed42, same anklet keyword) where no anklet was visible in full-body framing.
Experiment 2: Framing Changes (seed 789)
| F (close-up) | G (from below) | H (sitting) |
|---|---|---|
| | |
With seed 789, F showed no anklet, but G and H did. G showed the clearest gold charm anklet.
Summary
Experiment 1: Full-Body Shots
| Condition | Anklet Rendered | Positional Accuracy | Side Effects |
|---|---|---|---|
| A (Control) | None | - | - |
| B (anklet) | 0/3 | - | None |
| C (ankle bracelet) | 0–1/3 (unclear) | Unclear | None |
| D (gold anklet) | 0–1/3 (unclear) | Unclear | None |
| E (ankle chain) | 2/3 (chain rendered) | Waist, not ankle | Bikini color changed |
Experiment 2: Framing Changes
| Condition | Anklet Rendered | Positional Accuracy | Side Effects |
|---|---|---|---|
| F (close-up) | 2/3 | Correctly at ankle | Composition limited to feet |
| G (from below) | 3/3 | Correctly at ankle | None |
| H (sitting) | 2/3 | Correctly at ankle | Outfit shifted to dress |
Overall Analysis
Anklets are barely rendered in full-body compositions (Experiment 1: 0–2 unclear renderings out of 12). In contrast, changing the framing dramatically improves anklet rendering (Experiment 2: 7 out of 9 clear renderings).
These results suggest the cause is not “the model doesn’t understand anklets” but rather a spatial problem where the ankle is too small to render detail in full-body shots.
Practical recommendations:
- When you need a reliable anklet:
gold anklet+from below, legs focusis the most stable (3/3 success) - When close-up feet framing is acceptable:
close-up of feet and anklesalso works (2/3 success) - For sitting poses:
sitting, feet visibledoes produce anklets, but outfit specifications tend to be ignored
ankle chain was not tested with different framings, but given the body-chain behavior in full-body shots, other framings may also produce unintended placement.
Lab Director Comment: So the conclusion is that anklets not showing up is a rendering space problem, not a vocabulary problem. Change the framing and you get 7 out of 9.
gold anklet+from below, legs focusgoing 3/3 is genuinely practical. Still curious whether you can force anklets in full-body shots, but that’s a different experiment.
