From AdaGrad to RMSProp
Prof. Prabir Kumar Biswas, IIT Kharagpur · 2:35 · Study notes
Ask this session
Cloudinary under the hood8 Cloudinary URLs make this page. No render servers: each one is generated on request and cached.
Adaptive streaming
The player's HLS stream: Cloudinary encodes a ladder of renditions on demand and the player switches with the connection.
…/video/upload/q_auto/sp_hd_lean/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.m3u8
- q_auto
- AI-chosen quality: smallest file that still looks right
- sp_hd_lean
- adaptive streaming profile "hd_lean": a ladder of renditions the player switches between
Transcript (auto_transcription)
Word-timed transcript written by Cloudinary at upload. Pravaha indexes it for Find and Ask and uses it for subtitles.
…/raw/upload/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.transcript
Seek-bar previews
One sprite of frames for the thumbnails you see while scrubbing.
…/video/upload/q_auto/sp_hd_lean/fl_sprite/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.vtt
- q_auto
- AI-chosen quality: smallest file that still looks right
- sp_hd_lean
- adaptive streaming profile "hd_lean": a ladder of renditions the player switches between
- fl_sprite
- a sprite of frames for seek-bar thumbnails
AI highlights graph
Cloudinary's AI rates which parts are most interesting; the player draws it above the seek bar.
…/video/upload/e_preview,fl_getinfo/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c
- e_preview
- AI preview: finds the most interesting parts of the video
- fl_getinfo
- returns data about the video instead of the video
Poster frame
A frame 10% in, cropped to 16:9 around the speaker by AI, in the best format for your device.
…/video/upload/so_15.5,c_fill,ar_16:9,w_640,g_auto/f_auto,q_auto/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.jpg
- so_15.5
- starts at 15.5 s
- c_fill
- crops to fill the frame exactly
- ar_16:9
- aspect ratio 16 : 9
- w_640
- 640 px wide
- g_auto
- AI picks the focus (the speaker or slide), not a blind centre crop
- f_auto
- best format for each device (e.g. AV1, WebM, MP4, WebP)
- q_auto
- AI-chosen quality: smallest file that still looks right
Link preview card
The image WhatsApp, LinkedIn and Slack show for this page: a frame, a fade and text layers, all in one URL.
…/video/upload/so_15.5,c_fill,w_1200,h_630,g_auto/e_gradient_fade:symmetric_pad,y_-0.5,b_black/l_text:arial_30_bold:Pravaha,co_white,b_rgb:0f766e/fl_layer_apply,g_north_west,x_60,y_56/fl_layer_apply,g_south_west,x_60,y_70/f_jpg,q_auto/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.jpg
- so_15.5
- starts at 15.5 s
- c_fill
- crops to fill the frame exactly
- w_1200
- 1200 px wide · and 1 more like it
- h_630
- 630 px tall
- g_auto
- AI picks the focus (the speaker or slide), not a blind centre crop
- e_gradient_fade:symmetric_p…
- fades the image edges so text on top stays readable
- y_-0.5
- -0.5 px from the edge
- b_black
- background black
- l_text:arial_30_bold:Pravaha
- text layer “Pravaha” · and 2 more like it
- co_white
- text colour white · and 1 more like it
- b_rgb:0f766e
- background #0f766e
- fl_layer_apply
- places the layer defined just before · and 2 more like it
- g_north_west
- anchored to the top-left corner
- x_60
- 60 px from the side · and 2 more like it
- y_56
- 56 px from the edge · and 2 more like it
- c_fit
- fits inside the box
- g_south_west
- anchored to the bottom-left corner · and 1 more like it
- co_rgb:99f6e4
- text colour #99f6e4
- f_jpg
- JPG format
- q_auto
- AI-chosen quality: smallest file that still looks right
A Moment (vertical short)
What "Share this moment" makes at 0:30: trimmed, cropped to 9:16 following the speaker, with word-timed captions burned in. The first open can take a few seconds while Cloudinary analyses the video.
…/video/upload/so_29,eo_44/c_fill,ar_9:16,w_720,g_auto/l_text:arial_46_bold:sum%20of%20the%20squares,co_white,b_rg…/fl_layer_apply,g_south,y_220,so_1.5,eo_2.2/fl_layer_apply,g_south,y_220,so_12,eo_13.7/f_auto:video,q_auto/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.mp4
- so_29
- starts at 29 s · and 9 more like it
- eo_44
- ends at 44 s · and 9 more like it
- c_fill
- crops to fill the frame exactly
- ar_9:16
- aspect ratio 9 : 16 (vertical, for Reels and Shorts)
- w_720
- 720 px wide · and 9 more like it
- g_auto
- AI picks the focus (the speaker or slide), not a blind centre crop
- l_text:arial_46_bold:sum%20…
- text layer “sum of the squares” · and 8 more like it
- co_white
- text colour white · and 8 more like it
- b_rgb:000000b3
- background #000000b3 · and 8 more like it
- c_fit
- fits inside the box · and 8 more like it
- fl_layer_apply
- places the layer defined just before · and 8 more like it
- g_south
- anchored to the bottom · and 8 more like it
- y_220
- 220 px from the edge · and 8 more like it
- f_auto:video
- best format for each device (e.g. AV1, WebM, MP4, WebP)
- q_auto
- AI-chosen quality: smallest file that still looks right
Session in 60 seconds
The Study Pack's highlight reel: 5 AI-picked moments stitched into one labelled video with fl_splice.
…/video/upload/so_0,eo_16.6,w_1280,h_720,c_fill/l_video:pravaha:1ba747ae-bdb0-43b5-bd11-8a0eda32a02c,fl_spl…/so_14,eo_31.9,w_1280,h_720,c_fill/fl_layer_apply/fl_layer_apply/f_auto:video,q_auto/pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c.mp4
- so_0
- starts at 0 s · and 4 more like it
- eo_16.6
- ends at 16.6 s · and 4 more like it
- w_1280
- 1280 px wide · and 4 more like it
- h_720
- 720 px tall · and 4 more like it
- c_fill
- crops to fill the frame exactly · and 4 more like it
- l_video:pravaha:1ba747ae-bd…
- another video (pravaha/1ba747ae-bdb0-43b5-bd11-8a0eda32a02c) as a layer · and 3 more like it
- fl_splice
- joins the next clip onto the end of this one · and 3 more like it
- fl_layer_apply
- places the layer defined just before · and 3 more like it
- f_auto:video
- best format for each device (e.g. AV1, WebM, MP4, WebP)
- q_auto
- AI-chosen quality: smallest file that still looks right