Skip to content

fix: --mode rejected on seedance_2_0_mini, audio default, 480p price,… - #15

Open
onedaydigital wants to merge 1 commit into
oso95:mainfrom
onedaydigital:upstream-bugfixes
Open

fix: --mode rejected on seedance_2_0_mini, audio default, 480p price,…#15
onedaydigital wants to merge 1 commit into
oso95:mainfrom
onedaydigital:upstream-bugfixes

Conversation

@onedaydigital

Copy link
Copy Markdown

… portability

Six small fixes, each verified against higgsfield CLI v1.1.23 or arithmetic.

  1. --mode std on seedance_2_0_mini is rejected outright:

    $ higgsfield generate cost seedance_2_0_mini --prompt test --mode std ...
    Error: Unknown params: mode
    Hint: Run: hf model get seedance_2_0_mini

    higgsfield model get seedance_2_0_mini has no mode param at all (unlike seedance_2_0, which has mode: std,fast). pipeline.md set VOPTS="--mode std --resolution 720p" for it, so the whole draft/previz tier fails at the first call. Also noted in the Step 4 model table.

  2. generate_audio defaults to TRUE, so "do not pass --generate-audio" means audio is on, not off:

    $ higgsfield model get seedance_2_0
    generate_audio boolean true

    Now passes --generate-audio false on dives and connectors, and the gotcha is reworded. This is the same trap the Kling row already documents for sound: on. (The old note said the flag "errors on seedance"; it takes a boolean value fine.)

  3. 480p Monid cell was the 4-second probe price, not an 8s dive. Every other cell reproduces the documented formula (wh24*s/1024) to the cent:

    720p 8s: 1280720248/1024 * $7/1M = $1.210 (doc: $1.21) ok
    1080p 8s: 1920
    1080248/1024 * $7.7/1M = $2.994 (doc: $2.99) ok
    480p 8s: 854480248/1024 * $7/1M = $0.538 (doc: $0.28) <-- off
    480p 4s: 854
    480244/1024 * $7/1M = $0.269 (= the probe)

    The old figure also made the 5s connector ($0.35) cost more than the longer 8s dive at the same resolution. Corrected to $0.54 / $0.34 in SKILL.md and pipeline.md.

  4. stat -f%z is macOS-only. On Linux it prints nothing, so sizeBytes is empty and the Monid /put body is malformed — the default backend's upload helper. Replaced with wc -c <.

  5. Connector encode iterated conn_*.mp4 in glob order, so at N>=11 conn_10 sorts before conn_2 and gets written as conn2.mp4 — silently mis-ordering the chain. Now iterates by index.

  6. .[]0.result_url typo in Step 2; the working expression used everywhere else in pipeline.md is .[0].result_url.

… portability

Six small fixes, each verified against higgsfield CLI v1.1.23 or arithmetic.

1. `--mode std` on seedance_2_0_mini is rejected outright:

     $ higgsfield generate cost seedance_2_0_mini --prompt test --mode std ...
     Error: Unknown params: mode
     Hint: Run: hf model get seedance_2_0_mini

   `higgsfield model get seedance_2_0_mini` has no `mode` param at all (unlike
   seedance_2_0, which has `mode: std,fast`). pipeline.md set
   VOPTS="--mode std --resolution 720p" for it, so the whole draft/previz tier
   fails at the first call. Also noted in the Step 4 model table.

2. `generate_audio` defaults to TRUE, so "do not pass --generate-audio" means
   audio is on, not off:

     $ higgsfield model get seedance_2_0
     generate_audio    boolean    true

   Now passes `--generate-audio false` on dives and connectors, and the gotcha
   is reworded. This is the same trap the Kling row already documents for
   `sound: on`. (The old note said the flag "errors on seedance"; it takes a
   boolean value fine.)

3. 480p Monid cell was the 4-second probe price, not an 8s dive. Every other
   cell reproduces the documented formula (w*h*24*s/1024) to the cent:

     720p  8s: 1280*720*24*8/1024  * $7/1M   = $1.210  (doc: $1.21)  ok
     1080p 8s: 1920*1080*24*8/1024 * $7.7/1M = $2.994  (doc: $2.99)  ok
     480p  8s:  854*480*24*8/1024  * $7/1M   = $0.538  (doc: $0.28)  <-- off
     480p  4s:  854*480*24*4/1024  * $7/1M   = $0.269  (= the probe)

   The old figure also made the 5s connector ($0.35) cost more than the longer
   8s dive at the same resolution. Corrected to $0.54 / $0.34 in SKILL.md and
   pipeline.md.

4. `stat -f%z` is macOS-only. On Linux it prints nothing, so sizeBytes is empty
   and the Monid /put body is malformed — the default backend's upload helper.
   Replaced with `wc -c <`.

5. Connector encode iterated `conn_*.mp4` in glob order, so at N>=11 conn_10
   sorts before conn_2 and gets written as conn2.mp4 — silently mis-ordering
   the chain. Now iterates by index.

6. `.[]0.result_url` typo in Step 2; the working expression used everywhere
   else in pipeline.md is `.[0].result_url`.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
@onedaydigital

Copy link
Copy Markdown
Author

Quick verification if useful — generate cost doesn't create a job or spend credits:

$ higgsfield generate cost seedance_2_0_mini --prompt test --mode std --resolution 720p --duration 8
Error: Unknown params: mode

$ higgsfield model get seedance_2_0_mini | grep -c mode
0 # vs seedance_2_0, which has mode std,fast

$ higgsfield model get seedance_2_0 | grep generate_audio
generate_audio boolean true

Checked against higgsfield CLI v1.1.23 (built 2026-08-08) on 2026-09-02.
The 480p arithmetic is in the commit message and needs no CLI at all.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant