AI Coding
Lyric motion graphics on local GPUs, DeepSeek vs GLM — DeepSeek V4.1 Flash 1017 vs GLM 5.3 Flash
Scoreboard2
- DeepSeek V4.1 Flash 1017DeepSeekDeepSeek HarnessAuthor’s pick
- Top; unquantized, TP4 on SGLang, 1M context; 12 rounds
- GLM 5.3 FlashZhipu AI (Z.ai)DeepSeek Harness
- Bottom; mixed-precision quantized, TP4 on vLLM, 512K context; 24 rounds
Prompt
Lyric motion graphics on local GPUs, DeepSeek vs GLMThe post doesn’t quote the prompt wording.
DeepSeek’s lyrics lined up almost from the start; GLM was faster at generation but its lyrics stayed out of sync after 24 rounds.
@yu_ichi_suzuki’s view (paraphrased)
DeepSeek V4.1 Flash 1017 and GLM-5.3-Flash, both run locally on four DGX Spark-class machines in DeepSeek Harness, write a beat-synced canvas motion graphic for the same song from the song, lyrics and a background clip.
The post, itemised
- Kind
- Side-by-side
- Task
- Lyric motion graphics on local GPUs, DeepSeek vs GLM
- Category
- Animations/SVG
- Mode
- Multi-turn
- Posted
- Video length
- 60.5 s
- Likes
- 2
- Views
- 402
- Bookmarks
- 2
- Sponsored
- No
NotesThe author notes this is not a same-prompt comparison: the fix requests during each run differed.
Related
Posted by @yu_ichi_suzuki on X. aivs.tv claims no ownership: it hosts a copy for fast playback and links to the post on X. Request removal


