RenderDrama

What a second of AI video costs, channel by channel

Ratings, criterion by criterion and tool by tool

Three ways into the same scores. Read a criterion to see how every tool handles one thing; read a tool to see all seven at once; read a pair to see where two tools actually diverge. As of 2026-09-12.

Three ways into the same set of scoresThe same scores are published three ways. A criterion page holds one question fixed across every tool. A tool page shows one product's seven scores together. A pair page sets two products side by side on the criteria where they diverge. Which one to read depends on how far a decision has already narrowed.How far has the choice narrowedNot at allRead a criterionOne question held fixedacross every tool, scoredon a published definition.Down to oneRead a toolSeven scores together,where an uneven profilebecomes visible.Down to twoRead a pairSide by side on thecriteria where the twoactually differ.Same scores, three entry points
Fig. 1 Averages hide disagreement, so the pair pages exist for the case where two means match and the underlying rows do not.

A criterion page is the most useful of the three when choosing, because it holds one question fixed. Script handling, cast consistency, assembly, audio, cost — each is scored on the same published definition for every tool, and the definition is printed at the top of the page rather than assumed.

A tool page is the one to read when a decision has already narrowed to one product. It shows the seven scores together, which is where an uneven profile becomes visible: a tool can be strong on the thing a demo shows and weak on the thing a series needs.

The pair pages exist because averages hide disagreement. Two tools with the same mean can differ by three points on the criterion that matters to a particular job, and that only shows up when the rows are set side by side.

1By criterion

  • Script handling — Twelve of sixteen tools publish nothing about text longer than a prompt.
  • Cast consistency — Almost every vendor claims consistency; the ratings turn on whether a mechanism is described.
  • Shot control — Seven of sixteen offer a prompt box and nothing more.
  • Picture quality — The flattest column on the rubric: nothing sits below three and six reach five.
  • Sound — Two entries generate speech with the picture in named languages.
  • Assembly — The most polarised column here: six entries publish nothing about joining segments, three document it as a priced step with everything to do afterwards.
  • Cost transparency — Two entries publish no usable rate at all and are left unrated.

2By tool

  • SceneMixer — The most even profile here, and no other row averages above four.
  • LTX Studio — Second-highest average and the most even profile after the leader, with one blank: the studio itself is priced only as monthly credit plans.
  • Hailuo (MiniMax) — Fives on picture and sound, ones on script handling and assembly: one of the widest spreads here, strong where generation happens and silent around it.
  • invideo AI — An assembly-first profile: a five for joining segments and a three for script handling, with generation itself rated in the middle of the field.
  • Morph Studio — A flat profile with no fives and no ones: everything sits between two and four, which is unusual in a table that is polarised nearly everywhere else.
  • Veo / Flow (Google) — Fives on picture and sound, a one on script handling.
  • CapCut — An editor rather than a generator: a five for assembly, ones for script handling and cast consistency.
  • Wan (Tongyi Wanxiang) — Fours on picture and sound with a one on assembly.
  • Sora (OpenAI) — A five on picture quality and a blank on cost transparency, with an announced end date for the interface most people used.
  • Kling AI — A five on picture quality with ones on script handling and assembly, and a four for a documented approach to keeping characters consistent.
  • PixVerse — A mid-table profile with no peaks: threes and fours across the generation columns, the usual floor on script handling, and nothing at all above four.
  • Dreamina — Picture quality at four with everything else in the lower middle, and a two for joining segments that sets it apart from its own editor sibling.
  • Runway — A five on picture quality and a two on cost transparency, with the workflow columns low.
  • Vidu — A four for keeping characters consistent, a two for sound, a one for assembly, and the flattest published resolution curve in the ledger behind it.
  • Luma Dream Machine — A five on picture quality against ones on script handling and assembly, and a two on cost transparency.
  • Pika — The lowest average in the table, with a four for cost transparency standing alone against ones on shot control, script handling and assembly.

3Head to head

  • CapCut and Luma Dream Machine — Averages within 0.5 of each other and 4 criteria where the two differ by at least two points, starting with cast consistency.
  • CapCut and Morph Studio — Averages within 0.2 of each other and 4 criteria where the two differ by at least two points, starting with cast consistency.
  • CapCut and Runway — Averages within 0.3 of each other and 4 criteria where the two differ by at least two points, starting with cast consistency.
  • CapCut and Veo / Flow (Google) — Averages within 0.2 of each other and 4 criteria where the two differ by at least two points, starting with cast consistency.
  • Hailuo (MiniMax) and invideo AI — Averages within 0.2 of each other and 4 criteria where the two differ by at least two points, starting with script handling.
  • invideo AI and Kling AI — Averages within 0.4 of each other and 4 criteria where the two differ by at least two points, starting with script handling.
  • CapCut and Hailuo (MiniMax) — Averages within 0.4 of each other and 3 criteria where the two differ by at least two points, starting with cast consistency.
  • CapCut and Kling AI — Averages within 0.2 of each other and 3 criteria where the two differ by at least two points, starting with cast consistency.
  • CapCut and Vidu — Averages within 0.5 of each other and 3 criteria where the two differ by at least two points, starting with cast consistency.
  • invideo AI and Morph Studio — Identical averages and 3 criteria where the two differ by at least two points, starting with shot control.
  • invideo AI and Sora (OpenAI) — Averages within 0.3 of each other and 3 criteria where the two differ by at least two points, starting with script handling.
  • invideo AI and Veo / Flow (Google) — Identical averages and 3 criteria where the two differ by at least two points, starting with script handling.

4Single ratings

  • SceneMixer, script handling — The single five in a column where twelve of the sixteen rated tools publish nothing at all about text longer than a prompt.
  • LTX Studio, script handling — A three, shared with one other tool, in a column where most of the field publishes nothing at all about text longer than a prompt box.
  • invideo AI, script handling — A three in the thinnest column on the rubric, on a row that is otherwise strongest at the two far ends of the job rather than the middle.
  • Morph Studio, script handling — The single two in this column: above the floor that twelve tools share and below the partial statements the next value up requires.
  • SceneMixer, cast consistency — The single five in the column that decides whether serial work is possible, and a rating about disclosure rather than about results.
  • CapCut, cast consistency — The single one in this column, and the reason is category rather than quality: generating characters is not part of the published product.
  • SceneMixer, shot control — One of two fives in a column where seven of the sixteen rated tools publish nothing at all beyond a free-text prompt box of their own.
  • LTX Studio, shot control — One of two fives in this column, and the clearest single thing this vendor documents anywhere on the seven criteria of the rubric.
  • Morph Studio, shot control — The single four in this column: above the prompt-box band, and below a shot list that the tool consumes and a buyer can edit before it renders.
  • Pika, shot control — The single one in this column, which means a prompt box and nothing published about structuring a shot before it renders.
  • Hailuo, sound — One of two fives in this column, earned by naming the languages it speaks rather than by claiming multilingual speech and stopping there.
  • Veo and Flow, sound — One of two fives in this column, on a row that also holds a floor on script handling and a low value on cost transparency.
  • LTX Studio, assembly — The single four in a polarised column where six tools publish nothing about joining and three document it as a priced step.
  • Veo and Flow, assembly — A three, shared with one other tool, in the emptiest part of a column that otherwise splits into documented pipelines and bare generators.
  • Morph Studio, assembly — A three, shared with one other tool, and the value that keeps this whole row flat rather than giving it either a peak or a floor.
  • SceneMixer, cost transparency — The single five in this column, earned by publishing both halves of a credit rate rather than by charging less than anyone else.
  • LTX Studio, unrated on cost — One of two blanks in this column.
  • Sora, unrated on cost — One of two blanks in this column, on a row whose per-second API figures are published in detail somewhere else entirely.

What each score means, and what leaves a criterion blank, is set out on the scoring page rather than restated on every entry.

Elsewhere on the ledger: Channels, Unpriced, Price shapes, What each channel publishes, One model, several sellers, Single ratings, Questions, Terms, Learn, Data. How a figure earns a line is set out on the method page.