Extend Continuum study through 30 seconds

This commit is contained in:
Morpheus
2026-10-14 16:31:35 +00:00
parent 9d71f9747d
commit a06f89d548
3 changed files with 18 additions and 7 deletions
@@ -39,9 +39,10 @@ These values are held constant while testing prompts and clip scaling:
- Tiled mode: enabled through `DUMAS H3 Continuum Second Pass`
The imported graph starts at two 4-second Continuum chunks. The test matrix
in [`tests.json`](tests.json) increases the number of chunks until the
20-second target is reached. No sampler, scheduler, CFG, upscale or tile
changes are included in this study.
in [`tests.json`](tests.json) increases the number of chunks through 20, 24,
and 28 seconds, then tests a six-chunk, five-second-per-chunk configuration
for the 30-second target. No sampler, scheduler, CFG, upscale or tile changes
are included in this study.
## Prompt strategy
@@ -53,6 +54,7 @@ structure and specificity of the action plan:
2. Detailed chronological beats with explicit continuity.
3. Detailed beats with camera direction and physical state tracking.
4. Full 20-second plan with five linked 4-second beats.
5. Longform 30-second plan with six linked 5-second beats.
## Test procedure
@@ -61,15 +63,17 @@ structure and specificity of the action plan:
`tiled_upscale=true` and reaches 1920×1088.
3. Repeat with the next prompt variant using the same 2 chunks.
4. Keep the best prompt fixed, then test 3, 4, and 5 chunks.
5. Stop if the server errors, OOMs, or runtime growth becomes disproportionate.
6. Record each result beside the matrix with prompt hash, chunk count,
5. Continue through 24s and 28s; only then attempt the 30s five-second-chunk
configuration.
6. Stop if the server errors, OOMs, or runtime growth becomes disproportionate.
7. Record each result beside the matrix with prompt hash, chunk count,
runtime, output dimensions, and a motion/continuity verdict.
## Files
- `workflow.json` — supplied ComfyUI API graph.
- `prompt.txt` — controlled base prompt for the study.
- `prompt-variants.json` — exact smoke-test prompt variants.
- `prompt-variants.json` — exact smoke-test and 30-second prompt variants.
- `variables.json` — immutable constants and allowed test variables.
- `tests.json` — ordered test matrix.
- `character-reference.png`, `scene-reference.png` — fixed references.
@@ -12,6 +12,10 @@
"cinematic-camera": {
"description": "Adds shot language, motivated camera movement, and physical action detail.",
"text": "[0-4s] Begin with a smooth backward dolly as @R2 enters from frame left, maintaining a medium shot and realistic foot placement. The character notices motion off-screen right, stops, and turns with focused surprise. [4-8s] Hardly change the established direction: side-track beside @R2 toward the opening, then arc to a three-quarter view as the character peers through it and raises a cautious hand. Maintain continuous body position, clothing, lighting, and environment."
},
"longform-30s": {
"description": "A six-beat prompt designed for six five-second clips and a 30-second target.",
"text": "[0-5s] @R2 enters from the left and crosses the environment cautiously as the camera tracks backwards. The character notices an unusual movement in the distance and stops. [5-10s] Continue from the exact stance. @R2 turns right, approaches a narrow opening, and signals caution with one raised hand while the camera side-tracks. [10-15s] Wind moves through the scene. @R2 shields their face, discovers a small object on the ground, crouches, and examines it without changing hands or clothing state. [15-20s] @R2 stands with the object, looks toward a distant landmark, and moves quickly as the camera arcs from profile to a frontal three-quarter view. [20-25s] The distant event becomes clearer. @R2 reaches a vantage point, reacts with focused surprise, and takes one controlled step back while maintaining screen direction. [25-30s] @R2 adopts a ready stance and protects the object as the camera makes a slow push-in. Preserve identity, anatomy, object placement, lighting, geography, and physical continuity throughout. End with natural movement rather than a frozen pose."
}
},
"promotion_rule": "Promote the prompt with the strongest identity, action coherence, and transition quality at comparable runtime; do not change fixed constants during this comparison."
@@ -21,6 +21,9 @@
{"id":"CT-03","purpose":"2-chunk explicit camera/action prompt","prompt_variant":"cinematic-camera","chunks":2,"chunk_seconds":4,"target_duration_seconds":8,"status":"planned"},
{"id":"CT-04","purpose":"3 chunks using the best smoke prompt","prompt_variant":"best-of-smoke","chunks":3,"chunk_seconds":4,"target_duration_seconds":12,"status":"planned"},
{"id":"CT-05","purpose":"4 chunks using the best prompt","prompt_variant":"best-of-smoke","chunks":4,"chunk_seconds":4,"target_duration_seconds":16,"status":"planned"},
{"id":"CT-06","purpose":"5 chunks, full 20-second target","prompt_variant":"best-of-smoke","chunks":5,"chunk_seconds":4,"target_duration_seconds":20,"status":"planned"}
{"id":"CT-06","purpose":"5 chunks, full 20-second target","prompt_variant":"best-of-smoke","chunks":5,"chunk_seconds":4,"target_duration_seconds":20,"status":"planned"},
{"id":"CT-07","purpose":"6 chunks, first extension beyond 20 seconds","prompt_variant":"best-of-smoke","chunks":6,"chunk_seconds":4,"target_duration_seconds":24,"status":"planned"},
{"id":"CT-08","purpose":"7 chunks, extended multi-clip continuity","prompt_variant":"best-of-smoke","chunks":7,"chunk_seconds":4,"target_duration_seconds":28,"status":"planned"},
{"id":"CT-09","purpose":"6 five-second chunks, 30-second target","prompt_variant":"longform-30s","chunks":6,"chunk_seconds":5,"target_duration_seconds":30,"status":"planned","risk":"longer physical chunks may increase continuity or VRAM pressure"}
]
}